How tree species fill geographic and ecological space in eastern North America

Abstract

Background and Aims Ecologists broadly accept that the number of species present within a region balances regional processes of immigration and speciation against competitive and other interactions between populations that limit distribution and constrain diversity. Although ecological theory has, for a long time, addressed the premise that ecological space can be filled to ‘capacity’ with species, only with the availability of time-calibrated phylogenies has it been possible to test the hypothesis that diversification slows as the number of species in a region increases. Focusing on the deciduous trees of eastern North America, this study tested predictions from competition theory concerning the distribution and abundance of species.

Methods Local assemblages of trees tabulated in a previous study published in 1950 were analysed. Assemblages were ordinated with respect to species composition by non-metric multidimensional scaling (NMS). Distributions of trees were analysed by taxonomically nested analysis of variance, discriminant analysis based on NMS scores, and canonical correlation analysis of NMS scores and Bioclim climate variables.

Key Results Most of the variance in species abundance and distribution was concentrated among closely related (i.e. congeneric) species, indicating evolutionary lability. Species distribution and abundance were unrelated to the number of close relatives, suggesting that competitive effects are diffuse. Distances between pairs of congeneric species in NMS space did not differ significantly from distances between more distantly related species, in contrast to the predictions of both competitive habitat partitioning and ecological sorting of species.

Conclusions Eastern deciduous forests of North America do not appear to be saturated with species. The distributions and abundances of individual species provide little evidence of being shaped by competition from related (i.e. ecologically similar) species and, by inference, that diversification is constrained by interspecific competition.

INTRODUCTION

Ecologists broadly accept that the number of species present within a region balances regional processes of immigration and speciation against competitive and other interactions between populations that limit distribution and constrain diversity (Ricklefs, 2004, 2008; Cavender-Bares et al., 2006, 2009; Kembel, 2009). Species richness is a regional attribute (Rosenzweig, 1995; Losos and Schluter, 2000; Fine and Ree, 2006; Wagner et al., 2014), in the sense that species are produced mostly by the evolutionary divergence of populations in isolation (allopatry), which requires spatial heterogeneity on scales sufficient to reduce gene flow for long periods. Although palaeontologists and evolutionary biologists have considered diversification at length (Simpson, 1953; Givnish, 1997, 2010; Losos et al., 1998; Foote, 2000, 2007; Schluter, 2000), little has been said until recently about limits to continued species production within regions, beyond an ongoing discussion of the filling of niche space (e.g. Hutchinson, 1959; Rosenzweig and Taylor, 1980; Emerson and Gillespie, 2008). The fossil records of many groups of organisms suggest long-term stability in the number of species, especially compared with estimated rates of species formation (Alroy, 2000; Jaramillo et al., 2006; Rabosky and Sorhannus, 2009). Although ecological theory has, for a long time, addressed the premise that ecological space can be filled to ‘capacity’ with species (MacArthur and Levins, 1967; MacArthur, 1970, 1972; Pacala and Tilman, 1994; Chesson, 2000), only with the availability of time-calibrated phylogenies has it been possible to test the hypothesis that diversification slows as the number of species in a region increases (Pybus and Harvey, 2000; Ricklefs, 2007; Rabosky and Lovette, 2008; Rabosky, 2009a, 2010, 2013). Even so, estimated rates of species extinction rarely approach estimated speciation rates (e.g. Stadler, 2011; Etienne et al., 2012; Jetz et al., 2012), as required by equilibrium in species numbers, and conflicting evidence about the existence of diversity constraints on speciation has precluded a general consensus. Moreover, mechanisms by which species richness within a region feeds back on the rate of species production within that region have, as yet, received little attention. One possibility is that increasing species richness intensifies interspecific competition and limits the number of species that can coexist locally, thereby restricting immigration from outside the region or preventing allopatric sister species produced within a region from achieving secondary sympatry (e.g. Pigot and Tobias, 2013).

This study examines the distribution and abundance of species of trees in the deciduous forests of eastern North America for evidence bearing on the filling of niche space and consequent diversity limitation. Because the tree flora of eastern North America has not been fully characterized in a phylogenetic context, and because much of the diversity within the region arose elsewhere, no attempt was made to estimate change in the rate of diversification from phylogenetic information. Incidentally, however, because the tree flora is probably completely known, the estimation of regional species richness was tested by extrapolating rank–abundance curves obtained from local plots, following the approach of ter Steege et al. (2013).

The North American forest plots described by Braun (1950) were ordinated by means of their species composition; the partitioning of the ordination space among forest types was quantified by discriminant analysis; and the relationship of the ordination axes to climate variation was determined by canonical correlation analysis. Here, how species fill ecological and geographic space was examined using three approaches: (1) taxonomically nested analysis of variance assesses the evolutionary lability of distribution and abundance (Gaston, 1998; Ricklefs, 2011) as an indicator of long-term stability of populations; (2) the relationship of the average distribution and abundance of species within taxonomic families to the number of species per family assesses the impact of competition between close relatives (Ricklefs, 2009); and (3) comparisons of the distances in ecological space between closely related (i.e. congeneric or confamilial) species, and the distances between more distantly related species, test for ecological filtering vs. competitive sorting (Cavender-Bares et al., 2004, 2009; Lortie et al., 2004).

The results of these analyses show that: (1) contemporary species distributions are closely associated with climate; (2) population attributes are labile, with most of the variance in distribution and abundance occurring between species in the same genus; (3) distribution and abundance are unrelated to the number of closely related species within the region, suggesting that competition is felt broadly among the species that make up the regional community; and (4) distances in ecological space do not differ significantly between closely related and more distantly related species, which supports neither ecological sorting nor competitive effects on species distributions. These results suggest that factors that determine population characteristics evolve rapidly compared with the rate at which isolated populations achieve genetic incompatibility, and that competition between related species does not constrain distribution and abundance, implying that competition is diffuse in this system. If these inferences are correct, the mechanism by which diversity is self-limiting would seem to act by reducing the average abundance of all species in the system with a consequent increase in the rate of species extinction. At this point, however, mechanisms potentially responsible for the variation in species abundance and distribution, as well as their influence on the rate of extinction, are not understood.

MATERIALS AND METHODS

Most of the analyses reported here are based on tables of species abundances in small forest plots detailed in E. Lucy Braun’s book, Deciduous forests of eastern North America (Braun, 1950), first published in 1950 and reprinted in 1972.

The ‘plots’ included a mixture of short transects and small areas; many were close together; they were not fixed plots; they differed in area; and identifications were not represented by vouchered specimens (see, for example, Braun, 1936, 1940, 1942). Incorporating additional published information, Braun (1950) presented inventories for 347 plots in 91 tables covering nine named Forest Regions: Mixed Mesophytic (MMF), Western Mesophytic (WMF), Oak–Hickory (OHF), Oak–Chestnut (OCF), Oak–Pine (OPF), Southeastern Evergreen (SEF), Beech–Maple (BMF), Maple–Basswood (MBF) and Hemlock–White Pine–Northern Hardwoods, the latter sub-divided into two divisions: Great Lakes–St. Lawrence (GLF) and Northern Appalachian (NAF) (see Fig. 1). Altogether, the plots included only about one-third of the species of the region, and many of these at very low frequency. The least represented species in Braun’s data set (e.g. Magnolia tripetala) were less than 1000th as frequent in the sample as the most abundant species.

Fig. 1.

Locations of the forest plots included in Braun (1950), with the forest type indicated by the colour of the symbol. Many of the plots were located in the same locality, and their symbols overlap. The few Oak–Pine Forest Region plots are not shown. Map courtesy of Jason Knouft.

Data for each species were presented by Braun as the proportion of the total number of stems in each plot; most of the tables indicated the total number of stems per plot (median 104·5; 25–75 % quartiles, 70–165 trees). Only species that reach the canopy or mid-canopy of forests were considered. Thus, shrubs are not included. After updating the species names and resolving some inconsistencies in names, the data set included 96 ‘species’ on 347 sites. The number of plots occupied per species varied from one (several species) to 228 (sugar maple, Acer saccharum), and the product of the number of plots and the average percentage of trees per occupied plot (i.e. overall abundance) varied from 0.4 to 4800 (beech, Fagus grandifolia). For further analyses, to reduce the effects of sparse sampling, only sites with ≥5 species, and species present on ≥8 plots, were included, leaving 293 sites and 48 species (see Fig. 1). A non-metric multidimensional scaling ordination was conducted in PCORD (http://home.centurytel.net/∼mjm/pcordwin.htm) based on the species composition of the forest plots, using the three dimensions recommended by the software and extracting scores for the species and the plots on the ordination axes.

To examine the general relationship between tree species distributions and climate variation within eastern North America, 19 Bioclim variables (Hijmans et al., 2005) were downloade for the geographical co-ordinates of each forest plot (www.worldclim.org/bioclim). These variables are based on monthly average temperature and precipitation, and they describe the mean, seasonal variation in, and interaction between, temperature and precipitation. Four of the variables were log₁₀-transformed to normalize their distributions (Bio3, Isothermality; Bio13, Precipitation of wettest month; Bio15, Precipitation seasonality; and Bio16, Precipitation of wettest quarter) and principal components were extracted from the data. Three principal components extracted 89·6 % of the information from the 19 Bioclim variables and were retained for analysis.

Ecologists are increasingly incorporating phylogenetic information in analyses of community structure, taking into account the influence of shared evolutionary history on the similarity of species with respect to ecologically relevant traits (Cavender-Bares et al., 2009; Kembel, 2009). However, I have chosen not to use phylogenetically informed analyses here because of the poor quality of information on evolutionary relationships of North American trees, and the difficulty in placing the results of such analyses into an intuitively transparent framework. Instead, I have relied on taxonomically informed analyses that contrast differences between pairs of species belonging to the same genus or family, with differences between more distantly related species. Although taxon level does not correspond to a consistent time horizon, as long as taxa are monophyletic, more closely related and less closely related species are clearly distinguished at taxon boundaries, and their comparison permits conservative testing of hypotheses framed in terms of shared evolutionary history.

Uncertainty of phylogenetic relationships among North American tree species can be seen, for example, in a comparison of maples (Acer spp.) in the phylogeny of North American trees used by Hawkins et al. (2014) with their placement in the definitive phylogenetic treatment of the genus by Renner et al. (2008). Hawkins et al. included all nine North American members of the genus. Except for the sister relationship of A. rubrum (red maple) and A. saccharinum (silver maple), which is recovered in the Hawkins et al. phylogeny, the closest relatives of all other North American Acer are in Asia. For example, A. circinatum (vine maple) and A. spicatum (mountain maple) are sister taxa in the Hawkins et al. phylogeny, but are not closely related, with A. circinatum imbedded in an Asian clade of eight species, in the Renner et al. phylogeny. According to Hawkins et al., A. pensylvanicum (striped maple) is sister to the A. circinatum/A. spicatum clade, but the Renner et al. phylogeny places A. pensylvanicum in a distant relationship. Hawkins et al. also bring A. negundo (box elder, Manitoba maple) and A. glabrum (Rocky Mountain maple) together as sister species, but they are distantly related with unresolved positions in the backbone of the Acer radiation at least 40–50 million years ago (Mya). To emphasize further the general problem by a more or less random example, a recent (January 2015) check of the TimeTree of Life (http://www.timetree.org/; Hedges et al., 2006) for the age of the node uniting Quercus alba (white oak) and Q. rubra (red oak), two of the most prominent species of oak in eastern North America, was returned with ‘No molecular data available for this query’. This is not a comment on the quality of Hawkins et al.’s important paper, or on efforts to incorporate phylogenetic information into ecological analyses, but simply a caution about the quality of some data used in phylogenetically informed analyses, particularly without having an intuitive feeling for the consequences of using inadequate information about relationships in such analyses (e.g. Maddison and FitzJohn, 2015; Rabosky and Goldberg, 2015).

Non-phylogenetic statistical analyses in this study were conducted with the Statistical Analysis System, v9.2 (SAS Institute, Cary, NC, UA).

Additional analyses are described in the Results section where they are applied.

RESULTS

Discriminant analysis based on non-metric multidimensional scaling (NMS) scores

The 293 plots based were ordinated on the abundance of 48 species of tree in the reduced data set (see the Materials and Methods) and a discriminant analysis was conducted of the ten forest types based on the three NMS scores for each plot (SAS Proc CANDISC). Not surprisingly, the forest types were generally separable on the NMS axes (Wilks’ lambda = 0·21, F = 21·4, d.f. = 27, 821, P < 0·0001; Fig. 2). Generalized squared distances between forest types represent the squared distances between their centroids in units of the combined variance of species distributions within each forest type along the vector that separates two forest types. Thus, a generalized squared distance of 4 represents two standard deviation units separating the means of the distributions of two forest types, or an overlap of approx. 5 % of species in each group. Of the 45 pairwise comparisons between forest types, only 11 had generalized squared distances <4 (in particular, the Beech–Maple, Maple–Basswood, Mixed Mesophytic and Oak–Hickory forests form a cluster that extends from the Midwest into the Ohio River Valley). Overall, however, the ordination suggests that most forest types were well separated on the basis of their tree species composition. Resubstitution of plots into forest types based on discriminant analysis (SAS Proc DISCRIM) revealed a 41 % error rate overall, with errors (i.e. plots assigned to alternative forest types) exceeding 50 % for Beech–Maple, Mixed Mesophytic, Oak–Chestnut and Oak–Hickory forests. In general, however, the forest types are distinctive and reflect the varied geographic distributions of tree species in eastern North America.

Fig. 2.

Distribution of forest plots coded by forest type (see Fig. 1) on the first two axes of a non-metric multidimensional scaling ordination of 293 forest plots.

Distribution of forest types and climate variables

A discriminant analysis of forest types was conducted with the first three principal components of the Bioclim variables. Generalized squared distances between forest types tended to be large overall, with only six of 45 distances being <4; resubstitution errors were only 33 % of the total. A canonical correlation analysis between the three NMS scores and the three Bioclim principal component scores showed a strong relationship between the two data sets (Wilks’ lambda = 0·37, F = 39·6, d.f. = 9, 699, P < 0·0001), accounting for 57·5 and 12·8 % of the variance on the first two canonical axes. This confirms what has been obvious to forest ecologists, namely that the distribution of forest trees is, to a large extent, associated with geographic variation in climate. Indeed, the distribution of designated forest types is more closely associated with climate variation than it is with the distributions of their associated species, as would be expected because forest types are geographically defined and tree species cross forest type boundaries independently.

Phylogenetic analysis and diversification

Inferences concerning negative feedbacks of diversity on rate of diversification, generally attributed to the filling of ecological space, are mostly based on slowing rates of diversification perceived in lineage-through-time (LTT) plots or related analyses based on reconstructed phylogenies (Rabosky and Lovette, 2008; Rabosky, 2009b, 2013, 2014). As explained in the Materials and Methods, no attempt was made here to produce a phylogeny for the species in the Braun plots or for those in the deciduous forests of eastern North America. Although one might employ such a phylogeny to infer the temporal pattern of diversification within a region, phylogenies restricted to the species in geographically circumscribed samples are unsuitable because species typically are not produced locally within a small area, and many descendants of the early nodes that are represented in the tree diversify further outside the region of interest and are thus excluded, causing the slope of the LTT plot to decrease towards the present (Cusimano and Renner, 2010; Cusimano et al., 2012; Price et al., 2014).

Difficulties in estimating rates of diversification from local samples of species were emphasized by an analysis of the phylogenetic relationships of North American maples (genus Acer: Sapindaceae) (Renner et al., 2008), in which the nine represented North American species are descendants of seven different disjunctions with east Asian sister lineages within east Asian clades. Thus, most of the diversity of Acer in North America was produced outside the continent and dispersed into the region from eastern Asia or Europe across high-latitude land bridges. Diversifying clades generally lack geographic limits, except perhaps on isolated islands or island archipelagos, and thus have limited utility in addressing local controls on diversification. Local phylogenetic overdispersion is thought to result from competitive exclusion of species which are too similar (Cavender-Bares et al., 2004, 2009; Emerson and Gillespie, 2008; Kembel, 2009), but it might partly reflect the geography of species formation. Moreover, it is unclear whether local phylogenetic overdispersion could be a consequence of competition resulting from diversification on a regional scale.

How many species of tree occur in a region?

This question is central to any analysis of diversity, but it is often difficult to establish a firm number in poorly explored regions. Because the tree species of eastern North America are probably completely known, it is an interesting exercise to test the method ter Steege and his colleagues (2013) used to estimate the unknown number of tree species in the Amazon Basin. E. L. Little’s field guide to the trees of eastern North America (Little, 1980), excluding southern Florida, includes 315 native trees, defined as woody plants ‘… with an erect perennial trunk at least 3 inches (7·5 centimeters) in diameter at breast height (4·5 feet or 1·3 meters), a definitely formed crown of foliage, and a height of at least 13 feet (4 meters)’ (p. 11).

In contrast to eastern North America, the trees of the Amazon Basin are incompletely known, by an unknown amount. ter Steege et al. (2013) addressed the problem of unsampled tree species in the Amazon Basin and Guinean Highlands of South America by extrapolating a plot of log abundance vs. rank to the point at which the nth postulated species is represented by a single individual (log 1 = 0). Their extrapolation was based on more than half a million individual trees in 1170 plots representing 4962 species. The extrapolated number of species represented by one or more individuals in the Amazon Basin exceeded 16 000, suggesting that many species remain to be discovered. Of course, such extrapolations depend on the applicability of the log abundance–rank relationship (Magurran, 1988).

The approach of ter Steege et al. (2013) to the trees of eastern North America was applied based on the data of Braun (1950) and on an additional data set compiled by Xing et al. (2014). These latter authors analysed 4760 individual approx. 0·07 ha plots in 280 grid cells of 1 × 1 degrees latitude and longitude in the eastern USA (US Forest Service Forest Inventory and Analysis Program; FIA, http://fia.fs.fed.us/), which contained 31 729 individual trees representing 175 species. To bring the data of Xing et al. into line with those of Braun (1950), the Gulf States and the Atlantic Coastal Plain and Florida sites (sub-regions 12 and 13), which are dominated by pines and evergreen broad-leaved trees, were not counted here, reducing the count to 27 766 individuals representing 163 species.

ter Steege et al. (2013) estimated the total number of trees in their study area as the product of the local density of trees on the sample plots (565 ha^–1) and the area of the region, arriving at 3·9 × 10¹¹ individual trees, and estimated the abundances of the sampled species by extrapolating from the local plots. For the Braun data, the average proportion of the 48 most common species of tree on the 293 forest sites included in the analysis was calculated. I estimated the total number of trees in eastern broad-leaved, deciduous forests from the total extent of all forests in the region—about 384 million acres, of which 74 %, or 284 × 10⁶ acres (115 × 10⁶ ha), include predominantly broad-leaved, deciduous species (http://www.nationalatlas.gov/articles/biology/a_forest.html). Three temperate, North American CTFS forest inventory plots (SCBI, Haliburton, Wabikon: http://www.ctfs.si.edu/) have an average of 488 trees >10 cm diamter at breast height (dbh) per hectare. Extrapolated to the total area of forest in hectares, this comes to 488 × 115 × 10⁶, i.e. approx. 56 billion (10⁹) trees. The proportions of the 48 species of common tree ranged from 16·4 % (beech, F. grandifolia) and 16·1 % (sugar maple, A. saccharum) down to 0·03 % (sourwood, Oxydendron arboretum), translating to a range of abundances from 8·2 × 10⁹ down to 0·016 × 10⁹ individuals. The exponential relationship among the 48 species between abundance and rank was ln(individuals) = 22·476 (0·057 s.e.) – 0·1154 (0·0020 s.e.) rank (F_1,46 = 3296, P < 0·0001, R² = 0·986). Extrapolated to a single individual, i.e. the intercept of log abundance on the x-axis of species rank, this relationship predicts that the rarest species would be rank 194.8 (Fig. 3), which is 140 species (44 %) fewer than the known number.

Fig. 3.

Exponential decline in the estimated number of individuals per species of tree in deciduous forests of eastern North America as a function of rank from most to least abundant. The inset shows the extrapolation of the regression line to an abundance of one individual, presumably representing the rarest species in the region. Analysis based on the forest plots of Braun (1950).

In the compilation of Xing et al. (2014), tree density was approx. 95 individuals >12·7 cm dbh ha^–1, providing a lower estimate of approx. 11 billion trees in eastern US deciduous forests [sub-regions 12 (Gulf) and 13 (Coastal Plain and Florida) excluded to match Braun’s (1950) coverage]. The forest plots included 27 766 individuals of 163 species, the most abundant of which (red maple, A. rubrum) made up the proportion 0.0695, or approx. 764 × 10⁶ trees. The exponential relationship among the 163 species between abundance and rank was ln(individuals) = 20·232 (0·046 s.e.) – 0·0461 (0·005 s.e.) rank (F_1,161 = 9123, P < 0·0001, R² = 0·983). Extrapolated to a single individual, this relationship predicts that the rarest species would be rank 438.9, about 39 % above the number of species in eastern deciduous forests. These two exercises demonstrate that one should be cautious about attempting to estimate species richness within a region from rank–abundance plots constructed from limited sampling. Discrepancies of −44 % and +39 % from different sampling programs, compared with the known value of 315 species in the deciduous forests of eastern North America, indicate the potential error in applying such analyses.

Extrapolations based on rank–log abundance distributions depend on a number of assumptions that are generally violated by most sampling programs. Most importantly, the relative abundance of trees must actually conform to the chosen underlying distribution (Magurran, 1988), which is difficult to ascertain when most of the distribution has not been sampled. Moreover, individuals within the region must be sampled randomly. However, because data on abundance are typically extracted from a limited number of local plots, within which species typically are aggregated with respect to the entire region, the relative abundances of species present on the plots tend to be exaggerated regionally. Moreover, plots also tend to be aggregated spatially, which is particularly the case for Braun’s North American plots. This makes the decline of abundance with rank appear shallower than it is in the region as a whole, which in turn predicts a higher rank (i.e. number of species) when abundance is extrapolated to one individual. Finally, extrapolating relative abundance in plots to entire regions fails to account for the sampling distribution of abundance. As species become less common within the region, the abundances of the rarer species included in samples are overestimated simply because other species with similar regional abundance fail to be included by chance. This also has the effect of making the regional relationship of log–abundance to rank less steep than it actually is. Thus, using extrapolation to estimate the number of species within a region should be undertaken with caution.

How conservative are species distributions?

This question addresses how rapidly distribution and abundance within a region change over evolutionary time. Lacking a time-dated phylogeny for all the species in Braun’s study, an explicitly phylogenetic approach to trait evolution has not been taken here (cf. Blomberg et al., 2003; Swenson and Enquist, 2007; Revell et al., 2008; Crisp and Cook, 2012; Münkemüller et al., 2012). Rather, nested analysis of variance was used to estimate the proportion of the total variance in tree distribution that occurs between species within genera, genera within families, and among families (Gaston, 1998). When most of the variance resides among species within genera, one can infer that traits are labile over periods of time on the order of intervals required for species formation. To some extent, variance among congeneric species can reflect asymmetries in the partitioning of geographic ranges between allopatric sister species (Barraclough and Vogler, 2000); however, it also reflects expansion and contraction of populations within the time frame of species formation (Ricklefs, 2011). In the case of the Braun data, a nested analysis of variance (SAS Proc NESTED) was computed and individual taxon-level effects were tested (SAS Proc MIXED) for the number of plots, the sum of the local proportional abundance of a species over plots and the standard deviations of the distributions of species with respect to the NMS scaling scores of the individual plots on three axes. Most of the variance in these metrics occurred between congeneric species. The only significant variance components above the level of species-within-genus were for the sum of the local proportions among genera (66 % of the variance; P = 0·007) and the standard deviation of the species on NMS axis 2 among families (52 %; P = 0·006). Thus, although some evidence suggests conservatism in population traits above the species level, the signal is weak overall.

Does competition constrain species abundance and distribution?

Without experiments, it is difficult to test the effects of competition on populations (e.g. Cahill et al., 2008). The use of local resources by each individual tree undoubtedly reduces the resources available to others and constrains the total number of local competitors. An indirect test for competitive effects on abundance and distribution follows upon Darwin’s (1859, p. 110) insight ‘… that it is the most closely-allied forms … [which] generally come into the severest competition with each other; consequently, each new variety or species, during the progress of its formation, will generally press hardest on its nearest kindred, and tend to exterminate them.’ One prediction from this hypothesis of competitive effect in relation to degree of relationship is that the average of the distribution and abundance of individual species should decrease with increasing numbers of close (i.e. confamilial or congeneric) relatives (Ricklefs, 2009). Although area of distribution is sensitive to how recently a species has originated through allopatric divergence (Barraclough and Vogler, 2000), the spread of a species throughout a region is thought to be constrained by competition with close relatives (Pigot and Tobias, 2013). Potential rates of geographic spread and change in local population density are sufficiently rapid compared with the time scale of species formation, that the historical legacy of speciation events is probably obscured by stochastic variation (Pigot et al., 2012) and by contemporary interactions of a species with other species and the physical environment (Pigot et al., 2010; Pigot and Tobias, 2013).

In the Braun data set, the number of species per family distributed over the plots varied from eight (Fagaceae) and seven (Juglandaceae) down to 3–4 species in the Betulaceae, Pinaceae, Sapindaceae, Magnoliaceae, Oleaceae and Ulmaceae, and one species in the remaining 12 families. Linear regression (SAS Proc GLM) was used to test the relationships between the average number of plots per species in relation to the regional number of species within each family, for which the estimate was positive (rather than the expected negative relationship), but not significant (P = 0·11). The same analyses for the average local abundance per plot (positive, P = 0·13), and for the standard deviations of the distributions of each species’ occurrences on the three NMS axes (1, positive, P = 0·24; 2, negative, P = 0·16; 3, positive, P = 0·19) provide no support for the hypothesis that competition with close relatives within the region as a whole constrains the average distribution and abundance of tree species. Consistent with these results for Braun’s data, a similar analysis based on the data of Xing et al. (2014) [all regions; 175 species in 30 families; up to 33 species per family (Fagaceae)] produced an additional positive, but insignificant (P = 0·7) relationship between the number of individuals of each species recorded in all plots within the region and the number of confamilial species (Fig. 4). Thus, while trees undoubtedly compete, it is evident that species do so over a broad playing field including most of the other species beyond closer, and what one might have presumed were ecologically more similar, relatives.

Fig. 4.

The average number of individuals per species within families as a function of the number of individuals per family, based on the compilation of forest inventory analysis data of Xing et al. (2014). The relationship is slightly, but not significantly, positive.

Another prediction based on competition is that closely related species should occupy divergent distributions because they share adaptations for resource use and compete more strongly among themselves than among other, more distantly related species in the regional community (space, or niche, partitioning). Alternatively, related species might occupy similar distributions if their occurrences were constrained by shared adaptations to conditions of the physical environment (ecological sorting). Previous analyses have addressed this question by testing for phylogenetic overdispersion (i.e. greater ecological distance among closer relatives) (Cavender-Bares et al., 2006, 2009; Kembel, 2009); however, due to a lack of a suitable phylogeny for the sample of tree species, a taxonomically based comparison was adopted here. Specifically, the Euclidean distance on the three NMS axes, derived from the lineages × plots matrix, was calculated for each pair of the 48 more common species (n = 1128 pairs), and distances between species pairs in the same families or genera and those in different families or genera were compared. In the family-level comparison, the average distance between species in the same taxon (0·74, s.d. = 0·35, range = 0·084–1·69, n = 75) was less than the average distance between species in different taxa (1·01, s.d. = 0·44, range = 0·051–2·61, n = 1045) (G_adj = 17·0, d.f. = 8, P = 0·030) (Fig. 5). Similarly, in the genus-level comparison, the average distance between species in the same taxon (0·69, s.d. = 0·38, range = 0·084–1·69, n = 38) was less (but not significantly less) than the average distance between species in different taxa (1·01, s.d. = 0·44, range = 0·051–2·61, n = 1082). Thus, not only do related species appear to have a negligible effect on each other’s abundances and distributions (see above), they also tend to occupy more similar parts of the overall ecological space, which would presumably intensify competitive interactions between them.

Fig. 5.

Distribution of the Euclidean distances between pairs of species in the space defined by three non-metric multidimensional scaling axes of the reduced Braun (1950) data set. Pairs of species within families are marginally significantly closer on average than pairs of species in different families, as assessed by a contingency test of the expected distribution of within-family species pairs based on the between-family distribution.

Is local abundance associated with the breadth of ecological distribution?

A persistent theme in community ecology, formalized by McNaughton and Wolf (1970) and Brown (1984), has been that ecological distribution and local abundance have a positive relationship to each other. One might expect a positive relationship if the same features of the environment limited both local abundance and distribution across ecological gradients. Evidence bearing on this relationship is equivocal (Ricklefs, 1972; Gaston and Blackburn, 2000; Gaston et al., 2000; Blackburn and Gaston, 2001). Indeed, local abundance and regional distribution might plausibly be determined by different factors in the environment, perhaps the ability to extract limiting resources, on one hand, and tolerance of environmental conditions, on the other. This relationship was examined among species of trees in the eastern deciduous forests by analysing the relationship between the average proportional presence of each species in occupied plots (local abundance) and regional ecological distribution expressed as standard deviations of the distributions of the species across plots on the NMS axes, which are strongly correlated with climate variables, and also with geographic distance.

As shown in Fig. 6, the breadths of the ecological distributions of tree species are independent of the local abundances of the species on plots where they occur. This absence of a relationship is consistent with ecological distribution and local population density being influenced by different factors, rather than reflecting single limiting factors that vary in parallel within and between areas.

Fig. 6.

Relationship between the distribution of species of tree on non-metric multidimensional scaling axes 1–3 in relation to the average local density of each species on plots where it occurs. None of the relationships is significant (all P > 0·05, r² < 0·05).

DISCUSSION

Any theory of forest composition has to accommodate four observations represented by analyses described here: (1) abundance and distribution show little evidence of evolutionary conservatism, with most of the variance in these measures occurring between species within the same genus; (2) abundance and distribution exhibit no evidence of competition between close relatives; (3) ecogeographic centroid positions of species in NMS space are marginally closer than, or more similar to, those of species in the same family and genus than they are to unrelated species, suggesting some degree of niche conservatism; and (4) ecological distribution and local abundance appear to be constrained by different factors. Thus, species of forest trees within the broad-leaved deciduous forest region of eastern North America show little evidence of either ecological sorting or competitive constraint, consistent with other analyses (Cahill et al., 2008; Fritschie et al., 2014).

This is not to say that all species are demographically equivalent, in the sense of the neutral theory of biodiversity (Hubbell, 2001), or that ecological conditions have no influence on distribution and abundance. Clearly, trees compete for light, water, and soil nutrients, and it is likely that the number of tree individuals – or, at least, their metabolically active biomass – in the entire forest region is established by resource availability. However, little evidence supports the hypothesis that competition for resources limits the distribution of individual species in deciduous forests of eastern North America. Moreover, if competition were to limit the number of species in the regional community, it would have to do this through general reduction of species abundances and a resulting increase in the probability of species extinction in apposition to increasing species numbers through immigration and within-region speciation.

Any interpretation of diversity in North American deciduous forests is complicated by climate change that occurred during the late Pleistocene. Shifting temperatures and precipitation over the region created unique local climates and associations of trees species that are no longer present (Jackson et al., 1997, 2000; Jackson and Overpeck, 2000). The extinction of at least one formerly widespread species of Picea in North America has been recorded (Jackson and Weng, 1999). The situation in North America was undoubtedly less dramatic than the widespread extinction of European tree species following late Cainozoic climate cooling (Svenning, 2003), but the region might be experiencing similar delays in the return of species ranges to equilibrium with the contemporary climate (e.g. Normand et al., 2011). Thus, variations in the distribution and abundance of species might have, in part, historical roots. Given the lack of evidence that distribution and abundance have been influenced by competition, or that they reflect ecological sorting based on conservative adaptations to the environment, the impacts of climate change possibly have been profound, albeit largely unknowable (Veloz et al., 2012; Blois et al., 2014). Moreover, these impacts might have contributed to the apparent lack of structuring forces in the deciduous forests of eastern North America.

Maximum distances of species displacement in North America in response to climate change since the last glacial maximum were of the order of a few thousand kilometres, generally less, and landforms provided few obstructions to movement. Thus, extension of plant distributions of the order of a few hundred metres per year could be sufficient to keep species in equilibrium with changing climates, and this is perhaps not beyond the range of possibility (Clark et al., 1999; Svenning and Skov, 2007).

What causes variation in species abundance and distribution? In the absence of rapid climate change leading to ecological disequilibrium, one has to consider factors that create marked discrepancies in distribution and abundance. If tree species are interacting on a relatively level field, from a competitive standpoint, adaptation to local climate and edaphic conditions notwithstanding, it would seem that variation in populations might be related to interactions with herbivores, pathogens or other antagonists (van der Putten et al., 2001; Bever, 2003; Reinhart et al., 2005; Inderjit and van der Putten, 2010; Ricklefs, 2010a, b). Because variation in distribution and abundance occurs primarily between closely related species, this variation is probably linked to antagonists that specialize on individual species. Certainly, introduced pathogens and herbivores of trees, such as Dutch elm disease (Brasier, 1983), chestnut blight (Nuss, 1992) and the emerald ash borer (Poland and McCullough, 2006), are capable of wreaking havoc on populations over large areas. Recently, ecologists have documented the role of pathogens, especially soil fungi, having primarily intraspecific effects, in limiting populations of forest trees in both tropical (Comita et al., 2010; Mangan et al., 2010) and temperate regions (Johnson et al., 2012). The basic observation is that both local and regional densities of individual species of trees are inversely related to the impact of conspecifics on the survival and local abundance of seedlings. This is the so-called ‘Janzen–Connell effect’ (Janzen, 1970; Connell, 1971, 1978), whereby self-limitation through specialized pathogen pressure prevents individual species from dominating local communities and allows ecologically similar species to coexist. Whether this mechanism actually limits local or regional species richness is unresolved. However, one can incorporate host–pathogen coevolution to create variance in the host–pathogen balance among closely related host species (Pimentel, 1968; Ricklefs and Cox, 1972) and to explain non-random variation in population size and distribution in the regional community. That is, the balance between a host and its pathogen shifts as one or the other acquires new mutations that influence their relationship, or as pathogens are impacted by their own pathogens (Brasier, 1983; Nuss, 1992).

Deciduous forests of eastern North America appear to be open to the invasion and spread of new species, whether these are the products of species formation within the region or colonization of the region from outside. This openness of ecological communities is consistent with the invasion literature more generally (Sax et al., 2002, 2005). In addition, the taxonomic composition of eastern deciduous forests appears to have been stable over long periods, at least at the genus level (Latham and Ricklefs, 1993a) judging from the fossil record; most of the late Tertiary fossil genera continue to exist within the region (Eiserhardt et al., 2015). Similar forests developing under comparable ecological conditions in eastern Asia support many more species of tree (Latham and Ricklefs, 1993b; Qian and Ricklefs, 1999, 2000; Zhao and Fang, 2006), potentially because of higher rates of diversification (Xiang et al., 2004). Thus, in principle, tree diversity in deciduous forests of eastern North America could increase under greater speciation and immigration pressure, causing a compensating decrease in the average geographic extents and population sizes of individual species.

Whether regional species richness has reached a steady-state level cannot be determined from present-day distributions of species within the region. The distribution of node ages in a time-calibrated phylogeny of species in the region (Pybus and Harvey, 2000) also might not be helpful because a large proportion of the ancestral lineages arose outside the region. Thus, the question of diversity-dependent limitation of diversification is not readily answered for forest trees of eastern North America beyond saying that the region probably is not closed to the spread of new species, whether they originate from within or from outside the region. Analysis of current distributions of species informs us, in general terms, about processes that influence distribution and abundance, which seem to focus on evolutionarily labile interactions between host species and their pathogens. Whether these interactions in turn influence diversification, and whether their character changes with the diversity of species within the region, remains to be seen. Nonetheless, examination of the distribution and abundance of species within the regional setting emphasizes the dynamic nature of the regional community.