The number of tree species on Earth.

One of the most fundamental questions in ecology is how many species inhabit the Earth. However, due to massive logistical and financial challenges and taxonomic difficulties connected to the species concept definition, the global numbers of species, including those of important and well-studied life forms such as trees, still remain largely unknown. Here, based on global ground-sourced data, we estimate the total tree species richness at global, continental, and biome levels. Our results indicate that there are ∼73,000 tree species globally, among which ∼9,000 tree species are yet to be discovered. Roughly 40% of undiscovered tree species are in South America. Moreover, almost one-third of all tree species to be discovered may be rare, with very low populations and limited spatial distribution (likely in remote tropical lowlands and mountains). These findings highlight the vulnerability of global forest biodiversity to anthropogenic changes in land use and climate, which disproportionately threaten rare species and thus, global tree richness.

In 1994, Robert May (1) provided the optimistic observation that, by 2044, we would roughly know the current number of species on Earth. Half of that time period has already lapsed, and we are still far from that goal. Even for trees, which are among the largest and most widespread organisms on the planet (2–6), provide a wealth of ecosystem services for humans (7–9), and support much of terrestrial biodiversity (10), we still lack a fundamental understanding of how many species exist on our planet (3, 4, 11–13).

A growing body of evidence highlights details and mechanisms regarding the biogeographic patterns in tree species diversity, such as the number of species increasing consistently toward equatorial regions (14–16). With a manageable number of taxa, tree species in the higher latitudes are relatively well characterized. However, if hyperdominance of a small fraction of species in the tropics (17) is a general phenomenon, it would mean that these regions generally harbor a very large number of rare species, many of which are endemic. The contribution of rare species to ecosystem services may be relevant and is a topic of active research (18, 19), but it is challenging as most remain poorly documented (20–26). Therefore, estimating the number of tree species is essential to inform, optimize, and prioritize forest conservation efforts across the globe. Knowing diversity’s extents will be useful in several ways. First, it can help us to infer the evolutionary mechanisms that have generated diversity, so that we can predict how those same mechanisms may play out in the future. Second, it may assist in assessment of which systems may be most resilient to global change. Third, if undetected species are mostly rare and rare species are more vulnerable to extinction risk, having a better grasp of those numbers is essential to managing for biodiversity preservation. Finally, with an understanding of total species pools, it is possible to quantify the impacts of regional conservation efforts while also improving the ability to predict extinctions, manage diversity hotspots, or collect germplasm (22, 23).

Because of the limited extent of data available, estimates of tree species diversity in large geographic domains still rely heavily on expert opinions and compiled published lists of species descriptions that are geographically uneven in coverage (24, 25). Although local specialists have been increasingly joining efforts to consolidate species lists in many domains, these limitations have precluded efforts to scale this information to generate a global perspective. Here, based on a ground-sourced global database numbering ∼64,100 species [a value similar to a prior enumeration of the total of known tree species of ∼60,000 (17)], we developed estimates of the number of tree species at biome, continental, and global scales. Specifically, by comparing species accumulation curves (SACs) of tree species across different spatial scales, we estimated the number of species that have not been recorded in the global data compilation used herein.

Results and Discussion

Global-, Continental-, and Biome-Level Patterns.

We compiled a comprehensive global occurrence dataset with 9,353 (100- × 100-km) grid cell samples (called “samples” or “sampling units” hereafter of ∼1°) () by combining an abundance-based tree species dataset (the Global Forest Biodiversity Initiative [GFBI]) (Fig. 1), based on forest plots worldwide and comprising ∼38 million trees for 28,192 species, with a large high-quality occurrence-based dataset (TREECHANGE) that includes forest plots and botanical vouchers (26) (). It is important to note that despite the large number of grid cells, extensive data, and high mean global sample coverage (96.4%) (Table 1), sampling within grid cells in many regions of the world remains very sparse.

Fig. 1.

The number of tree species and individuals per continent in the GFBI database. This dataset (blue points in the central map) was used for the parametric estimation and merged with the TREECHANGE occurrence-based data (purple points in the central map) to provide the estimates in this study. Green areas represent the global tree cover. GFBI consists of abundance-based records of ∼38 million trees for 28,192 species. Depicted here are some of the most frequent species recorded in each continent. Some GFBI and TREECHANGE points may overlap in the map.

Table 1.

Observed, asymptotic, and adjusted tree species richness and sample coverage at continental and global scales (note that the global value is lower than the sum of the continental ones due to overlapping species among continents [Fig. 4] and due to independent estimators being run for each continent and globally)

Continent	No. of sampling units (∼1° grid cells)	Species (observed)	Sample coverage, %	Chao2 (asymptotic)	95% CI lower Chao2 (asymptotic)	95% CI upper Chao2 (asymptotic)	Chao2adj	95% CI lower Chao2adj	95% CI upper Chao2adj	To be discovered (∼Chao2adj − Sobs)	Hotspot biomes
Global	9,353	64,088	96.4	89,147	89,141	89,152	73,274	73,271	73,276	9,186
Africa	1,575	10,441	96.0	14,031	14,028	14,033	11,875	11,874	11,877	1,434	Tropical /subtropical moist and dry forests, mainly in the Congo River basin
Eurasia	2,896	14,071	94.3	18,311	18,305	18,316	16,264	16,262	16,265	2,193	Tropical/subtropical moist and dry forests, mainly in Southeast Asia
North America	2,418	8,646	98.6	10,295	10,290	10,299	11,131	11,129	11,134	2,485	Tropical subtropical moist and dry forests, mainly in Central America
South America	1,461	27,186	95.0	46,738	46,729	46,747	31,112	31,110	31,115	3,926	Tropical/subtropical forests, grasslands, savannas, and shrublands, mainly in the Amazon River basin and Andean high mountains
Oceania	1,003	6,680	97.4	9,273	9,267	9,277	8,235	8,232	8,237	1,555	Tropical/subtropical moist forests, mainly in northeast Australia and the Pacific Islands

We also list some biomes that are hotspots of undiscovered species in each continent (). Bold indicates the number of species to be discovered globally and continentally.

Fig. 4.

Species richness partitioning among continents. Estimates of the percentage of continental endemic (bold percentage values close to each continental map are based on the Chao2adj estimator) () relative to the estimated richness per continent and shared species among continents (numbers in overlapping sets). In the center (bold percentage values at the intersection of all sets), the percentage of shared species among all five continents is shown.

From this dataset, with a nonparametric estimator [Chao2 (27)], we calculated occurrence-based values of potential global and continental tree species richness (, Fig. 2, and Table 1). This estimator is sensitive to accurate quantification of the numbers of uniques and duplicates (below and ), and it is known that there are problems with false uniques in forest species richness datasets (24). These Chao2 values may thus represent an overestimate to the degree that tree species recorded in only one sampling unit have been mistakenly identified as unique. Therefore, we estimated the true number (Chao2adj) of unique species (28) () by accounting for the relationships of uniques, duplicates, triplets, and quadruplets to constrain the estimated numbers of unique species and from this adjusted number, computed a more conservative estimate of global tree richness, which is ∼73,300 species (Chao2adj). Based on the good performance of this estimator and its adjusted version reported in previous studies (9, 29–33), we considered the adjusted value (Chao2adj) our most reasonable approximation to global tree species richness. We then derived SACs at global (Fig. 2) and continental (Fig. 2) scales. Moreover, for each continent, from the observed number of tree species, we also estimated the asymptotic richness at the within-continent biome-level extent (Fig. 3 and ).

Fig. 2.

Occurrence-based accumulation curves at global (A) and continental (B) scales. In A, nonparametric (interpolated) and asymptotic (extrapolated) species numbers from Chao2 (upper–lower 95% CI as shaded areas around the means; note that the CI shaded area is narrow because of the high number of sapling units), the Chao2adj estimate for the true number of singletons (red line) vs. the number of samples (1° grid cell ∼100 × 100 km), and the number of species listed in GlobalTreeSearch (green line) are shown. In B, nonparametric (interpolated) and asymptotic (extr., extrapolated) estimates (upper–lower 95% CI as shaded areas around the means) and Chao2adj values for the true number of singletons (dashed lines) are displayed vs. the number of samples (1° grid cell ∼100 × 100 km) within continents; the percentage of the global estimated richness in each continent is shown in the cartogram in B, Inset (total richness per continent is reported in Table 1).

Fig. 3.

Biome-level tree species richness estimates. The map shows the number of tree species estimated (S estimated from Chao2adj) in terrestrial biomes of each continent as a color gradient from low richness (yellow) to high richness (red). More information is provided in .

At the global scale, we infer that there likely are ∼9,200 tree species yet to be discovered (Table 1), given the ∼64,000 species already encountered (3, 4, 34–37). Our estimates at continental scales (Fig. 2 and Table 1) show that roughly 43% of all Earth’s tree species occur in South America, followed by Eurasia (22%), Africa (16%), North America (15%), and Oceania (11%). However, a lack of saturation (driven by the existence of high numbers of species uncommon in the landscape, incomplete sampling, or both), particularly in the South American accumulation curve (Fig. 2), suggests that our estimates may still be incomplete accounts of continental and global tree species richness. More undiscovered species likely occur in South America than any other continent. Our findings are in general agreement with recent studies of Amazonian plant diversity, which suggested that there are many undiscovered species; moreover, different approaches to the problem arrived at different estimates of total numbers of known and unknown species, suggesting that, as a scientific community, we still have much more work to do to arrive at accurate estimates regionally, continentally, or globally (12, 24, 34–37). Additionally, a considerable number of species have likely not yet been encountered in each of the other four continents as well (Table 1), most likely in the species-rich and more poorly studied tropical regions within each (see below).

Our biome-level estimates of tree richness (Fig. 3 and ) provide a more detailed description of the distribution of species richness within continents and shed more light on South America’s extremely high total tree diversity. As expected, the highest estimates of tree species in all continents are for the tropical and subtropical moist forest biome; for example, roughly half to two-thirds of all already known species occur in these forests on all five continents (). Moreover, the hotspots of undiscovered species (Table 1 and ) may largely occur in these same species-rich and undersampled (37) regions. However, high numbers of known and unknown species also occur in other biomes, including tropical and subtropical dry forests, temperate forests, mangrove forests, and areas classified as nonforested biomes (e.g., lowland and montane grasslands, savanna, shrublands, deserts) but that include considerable areas of tree-rich, and often speciose, vegetation. The high total tree diversity in South America is dominated by the lowland wet tropics and subtropics, yet roughly one-third of all tree species on that continent are found only outside of that biome.

Rarity in Forests Worldwide.

We also calculated indices of tree species rarity (percentages of singletons and doubletons) at continental and global scales () to help illuminate possible within-sample and among-sample abundance patterns. In fact, since the sample coverage deficit (1 – coverage = slope of the SAC at its right-hand end) is a statistically rigorous way of assessing the incompleteness of sampling (38), the proportion of singletons/uniques is, thus, strongly driven not only by long tails on the underlying species abundance/occurrence distribution but also, by sampling intensity/completeness.

Our most reliable abundance-based asymptotic richness estimates depend on the total number of observed species and the number of species with only one (singletons) or two (doubletons) individuals in each sample (which may represent measures of abundance-based rarity). Similarly, occurrence-based estimates depend on the number of species present in only one (unique) or two (duplicate) samples of each continent (which may represent measures of occurrence-based rarity). Rarity data within samples (α; i.e., from the abundance-based dataset) provide an indication of the relative proportion of species that are rare at the landscape to small regional scale represented by individual grid cells (100 × 100 km). The global rarity value is 33%, with Africa (38%) and South America (37%) having the highest percentage of species rare within samples and North America (17%) and Eurasia (24%) having the lowest (). It is important to note that our data do not mean that one-third of all species occur only once or twice in nature; instead, their rarity in our dataset suggests their rarity in nature but with unknown distributions of real occurrences. The ratio of singletons to doubletons within grid cells is higher in Africa and Oceania followed by South America and is quite low in North America and Eurasia.

From the rarity data among samples (occurrence-based rarity), we estimated that South America accounts for the highest total number of rare species (∼8,200) followed by Eurasia (∼6,100) and Africa (∼3,900). In Eurasia and North America, the percentage of species rare among grid cells was ∼43%, and it was <40% in the other continents, with the lowest value in South America (∼30%) (). The ratio of singletons to doubletons among grid cells in North America (1.83) is the highest among continents; for all other continents, it is lower than 1.5. At a global scale, percentage abundance-based rarity is higher than occurrence-based rarity, while the ratio of singletons to doubletons shows the opposite trend. Since we were aware that the numbers (and proportions) of singletons/uniques and doubletons/duplicates (and their relative magnitudes) are very much a function not only of true rarity but also of sampling effort, in relation to true richness, we estimated all indices adjusting them for “true singletons/uniques” (). However, our findings still confirm that most forests are likely to be dominated by just a few tree species (17) and include a long tail of rare species, which represents a consistent 30 to 40% of the overall tree richness in all continents. Although more species-rich regions (such as South America and Africa) have higher abundance-based rarity, North America and Eurasia (which contain more of a mix of biomes) showed higher occurrence-based rarity, and this finding could provide insights to better understand the biogeography of tree species on Earth.

Overall, almost a third of global tree richness on Earth is made up of rare species, which appear only once or twice in our samples. Thus, if the global forest system is dominated by a relatively modest number of abundant tree species, the global number of tree species strongly depends on those rarely detected (∼35%) () and undetected species (some large fraction of the ∼9,200 unobserved over the ∼73,300 estimated) (Table 1) (34). These results highlight the vulnerability of global forest biodiversity to anthropogenic changes, particularly land use and climate, because the survival of rare taxa is disproportionately threatened by these pressures (16–19). The higher threats for rare species are an important concern if we consider that their functions in ecosystems, the services they provide, and the ecoevolutionary patterns of these hyperrare tree species are still poorly known (16–20, 25).

Comparisons across Continents.

To better understand the biogeography of richness patterns across land masses, we also estimated species turnover among continents (Fig. 4). Specifically, we combined the data of the five continents to obtain the values of estimated tree species richness in all 31 possible intersections (). The two continents that share the highest estimated numbers of tree species are North and South America (Fig. 4), which is not surprising since these continents are interconnected by land (since about 3 Mya) in a region where nearby species-rich tropical forests occur on both continents. Consistent with this pattern, the second-highest number of shared species is between Eurasia and Oceania (Fig. 4), which had a geological continuity through the Southeast Asian archipelago that is another hotspot of tree diversity. Overall, other than the highest number of rare species, South America also shows the highest estimated percentage (49%) of continental endemic species (Fig. 4), while Eurasia and Africa account together for almost another 32% of unique tree species in the world. The percentage of shared species estimated among all five continents is lower than 0.1 (Fig. 4).

To summarize our main findings, we estimated that the absolute number of tree species on Earth is considerably higher than previously reported, with 14.3% more species than currently known to science (3). By establishing a quantitative benchmark, this information could contribute to tree and forest conservation efforts and the future discovery of new trees and associated species in certain parts of the world. For instance, considering that we estimated that about 31,100 tree species are expected in South America (Chao2adj estimator) and those known to science are about 27,200 (Table 1), there might be about 3,900 tree species yet to be discovered in this continent, and most of them could be endemic (Fig. 4) and located in diversity hotspots of the Amazon basin and the Andes–Amazon interface. This makes forest conservation of paramount priority in South America, especially considering the current tropical forest crisis from anthropogenic impacts such as deforestation, fires, and climate change. Similar arguments can be made about the prioritization of conservation of tropical and subtropical forests on other continents given the considerable numbers of likely undiscovered species on each and their likely rarity. For example, there are likely high numbers of undiscovered species in Central America and in Southeast Asia.

This study accelerates our science by estimating global tree richness with a more extensive dataset and more advanced statistical methods than previous attempts. However, both the underlying data and Chao richness estimators and adjustments are imperfect. We recognize several methodological issues that might have potentially biased our estimates and/or contributed to uncertainty. The first involves the uneven and unrepresentative distribution of the sampling areas in the globe and within continents, which is an issue despite the high–sample coverage metrics that we used. The second involves the possibility that some species might have been misclassified due to misidentification, failure to update taxonomic name changes, and misspellings, which could reduce accuracy in estimates of species numbers (24, 33, 34). There is compelling evidence of errors in most biodiversity datasets due to the inclusion of false uniques (24). For example, if two botanists in different parts of the same forest region encounter the same species of rare and unfamiliar tree, they may identify it differently or use different synonyms to identify it, biasing the count of uniques and the estimators. Therefore, because of the likely discrepancy between the actual proportion of uniques in a sample and the observed unique count included in our datasets, we estimated the true number of unique species (28) and from this adjusted number, computed and focus on a more conservative estimate of global tree richness, which is ∼73,300 species (Chao2adj). There is also uncertainty about the accuracy of nonparametric estimators. Previous studies report that nonparametric estimators give lower values of tree species richness than parametric ones for the Amazon basin (34–36). However, our nonparametric estimate of tree species diversity in South American tropical forest biomes was higher than both parametric estimation and previous estimates in the Amazon (36). This might have resulted from previous studies being mainly based on Amazon lowlands, ignoring highlands. Thus, we examined sample completeness comparing continents but limiting their latitude to 23°N and S (tropical regions) (). Results generally showed similar sample coverage at the grid-scale size used.

Future estimates of tree species richness in tropical, subtropical, and montane areas on all continents will be more accurate if an increased sample size is obtained (37), especially from areas poorly investigated. This begs the question on why South America alone could harbor >40% of all tree species. Compared with forest ecosystems on other continents, South America could have offered a larger continuous tropical forest area, a higher rate of speciation, a more robust mechanism of biodiversity maintenance, and reduced extinction rates [for instance, mild climates and the shortest period of human disturbance (39, 40)]. We also noticed that the SAC of South America continued to rise along the samples, whereas those of other continents start to level off, supporting the idea that undiscovered species numbers are likely high there, including in the Andean forests between 1,000- and 3,500-m altitude. A key challenge now is to install more plots in the Amazon-Andean transition zones, and to identify and monitor the trees within these plots.

Overall, our study points toward an estimated global tree richness (∼73,300) that is roughly 14% higher than numbers known today (3, 4), with many unknown species belonging to the tail of rare ones and often endemic to certain regions all across the globe. These results highlight the vulnerability of global tree species diversity to anthropogenic land use changes and to future climate (16–18). Losing regions of forest that contain these rare species will have direct and potentially long-lasting impacts on the global species diversity and their provisioning of ecosystem services (18–20). These results demonstrate both the lack of knowledge we still have about the tree species within our global forest systems and the value of approaches to help fill those gaps, which will be useful in providing fundamental insights about the diversity of life on our planet and its needed conservation.

Materials and Methods

Dataset and Sample Coverage.

We used the tree definition agreed on by IUCN's (International Union for Conservation of Nature) Global Tree Specialist Group (GTSG): “a woody plant with usually a single stem growing to a height of at least two meters, or if multi-stemmed, then at least one vertical stem five centimeters in diameter at breast height.” A tree inventory abundance dataset from 105,749 forest plots, ∼38 million stems of 28,192 species, distributed across all five continents was compiled from the GFBI (https://gfbinitiative.net/) database. For the Tonga and Niue data in the GFBI dataset, the original source was the New Zealand National Vegetation Survey Databank. For the estimation of the total number of tree species worldwide, we further compiled an independent occurrence dataset that we combined with the GFBI data. The occurrence-based dataset (hereafter, TREECHANGE) consists of taxonomy and location of >6 million tree individuals. Being a major data infrastructure itself, this dataset represents species occurrence information and encompasses a huge variety of data—from ground-sourced forest plot data (similar to the GFBI). Supported by a large body of collaborating institutions all over the world, this dataset features extensive global coverage and has been used across many large-scale studies (26). A limitation of the TREECHANGE dataset is that its underlying datasets do not have a coherent and consistent design and sampling scheme, but as described below, it complements the calculation of the estimated total number of tree species worldwide based on GFBI data. We extracted taxonomic data and associated geographic coordinates from five main data aggregators of species occurrences: the Global Biodiversity Information Facility [accessed through rgbif R package (41)], the public domain of the Botanical Information and Ecological Network v.3 [accessed through the BIEN R package (42)], the Latin American Seasonally Dry Tropical Forest Floristic Network [DRYFLOR (43)], the RAINBIO database (44), and the Atlas of Living Australia [ALA; accessed through the ALA4 R package (45)]. The species list was initially extracted from a world tree species checklist [GlobalTreeSearch (46)]. We checked for taxonomic correctness using the Taxonomic Name Resolution online tool (47), following a quality assessment and control of the data using the workflow outlined in ref. 26. This workflow minimized common errors associated with occurrence data (43). GlobalTreeSearch uses the tree definition agreed on by IUCN's GTSG above mentioned.

For abundance-based analyses, we used the GFBI tree species dataset (at its original plot size), whose samples cover a total area of more than 73,000 ha (). Then, to perform occurrence-based estimations, we compiled a larger and more comprehensive global dataset with 100- × 100-km sampling units (∼1° grid cells) by combining the abundance-based data in the GFBI tree species dataset, which were converted in presence/absence occurrence data and pooled with the high-quality large occurrence-based TREECHANGE dataset. Globally, this yielded a dataset of 9,353 sampling units, with 696,063 occurrences. At the continental level, the combination of the two datasets to obtain a large occurrence-based dataset also yielded a number of sampling units somewhat comparable, in the sense of being a similar order of magnitude (Africa: 1,575; Eurasia: 2,896; North America: 2,418; South America: 1,461; Oceania: 1,003).

To ensure that our estimations of species richness were not biased by differences in sample coverage (e.g., an estimate of the total probability of occurrence of the species observed in the sample, taking into account species present but not detected) among continents, we estimated the inventory completeness (as defined by ref. 48) for the complete database and for each continent separately using the Chao–Shen sample coverage estimator (38, 48), which is a bias-reduced estimator of sample completeness:where f1 and f2 are the numbers of singletons and doubletons (for abundance-based data) or the species occurred in only one (uniques) and in two (duplicates) 100- × 100-km (∼1°) samples (for occurrence-based data), respectively; n is the total number of individuals (for abundance-based data) or occurrences (for occurrence-based data) in the sample; and C is the proportion of the total number of individuals (for abundance-based data) or occurrences (for occurrence-based data) in an assemblage (observed and not observed) that belong to the species represented in the sample (49, 50).

Because estimates of species richness can be strongly dependent on differences in inventory completeness, we checked whether sample coverage was similar in all five continents. Since all continents showed a similar proportion of sample coverage (all >94%), both from occurrence- (Table 1) and abundance-based data (), we confirmed that our global estimate—based on global sample coverage of 96.44% (occurrence data) and 99.97% (abundance data)—was not disproportionately influenced by any specific continent. However, the slightly lower occurrence-based sample coverage of South America and Eurasia, with 95 and 94.26%, respectively, and the clustered distribution of some plots could explain the nonsaturating trend of their accumulation curves compared with the other continents (Fig. 2). We also note that sample completeness at finer scales would be lower in all continents.

We selected the continental scale for our estimates, together with the common study frames of biomes (51), because nonparametric species richness estimators perform better when samples are collected in a continuous incremental area without relevant landmass separation such as oceans (31–33).

For instance, working at a global biome-level only would ensure that the current climatic conditions are similar, but this approach to estimate species richness, taken alone, would reduce the information implied in the estimates because they would be affected by several factors. 1) Within each across-ocean biome, there are still important ecological and evolutionary differences that would affect the estimates at the global biome level [in fact, conventional levels of ecological hierarchical organization are not scale dependent (52), whereas species richness estimates are]. 2) With nonparametric estimates based on SACs, it is better to ensure a continuity of sampled areas (e.g., continuous terrestrial lands) (53, 54). 3) The ecological conditions that have shaped the evolutionary patterns (phylogeny and diversity) of tree species on Earth were much different when continents were conglomerate in Pangea (55) and then slowly shifted away (i.e., during this long geological time, biomes were much different to current ones) (56–58).

Therefore, other than estimating global tree species richness at a global biome level (Fig. 3 and ), we analyzed continental richness to also account for evolutionary changes in response to the biome main variables (latitude, climate, solar radiation, etc.), which shaped current tree diversity. Adding the figures at a continental (and a continental biome) level, we ensured that our estimates are based on the 135-My biogeographical and temporal continuity of the five main vegetated landmasses, which is an implied assumption of the estimators. This approach also allows a better discussion of the results for species turnover among continents (Fig. 4), which might be a result of their connections in Laurasia and Gondwana and the following continental drift.

Species Richness Estimators.

We initially computed a parametric estimate of species richness on the abundance-based data for 28,192 species from the GFBI dataset (). In particular, we considered the Fisher’s α for abundance data (calculated from http://groundvegetationdb-web.com/ground_veg/home/diversity_index).

We found that the abundance-based Fisher’s α underestimated the absolute species richness because our global () Fisher estimate was close but lower than the observed number of species in our occurrence-based dataset (64,100 from GFBI + TREECHANGE). Because this parametric estimator assumes a log-series distribution of abundances (59), we performed a goodness-of-fit test and evaluated it with a Kolmogorov–Smirnov test of whether our global and abundance data fit a log-series distribution. Since all datasets (global: D = 0.1, P = 1; Africa: D = 0.1, P = 1; Eurasia: D = 0.5, P = 0.17; North America: D = 0.2, P = 0.99; Oceania: D = 0.2, P = 0.99; South America: D = 0.2, P = 0.99) follow a log-series distribution, we calculated the α-values. At a global level, we obtained a Fisher’s α-value of 3,040 ().

We used this value and the most recent estimates on the global number of trees by Crowther et al. (60) to estimate the global number of species from Fisher’s classical equation (61):where N is the total number of trees and α is the Fisher’s α-parameter. This yielded an estimate of 62,624 to 62,915 species (lower–upper bootstrap 95% CI) from the 3.04 ± 0.19 × 1012 (±95% CI) global tree stems calculated by Crowther et al. (60). Although Fisher’s parametric approach stands on the very strict assumption of infinite log-series species abundance distributions, giving rise to overestimation of hyperrarity (62, 63), it estimated slightly less than the observed number of species in our occurrence-based dataset (using the α-value derived from our abundance-based dataset). We thus did not further employ this estimate. Instead, with the larger occurrence dataset composed of GFBI (converted to presence/absence) and TREECHANGE data, we then calculated the Chao2 index, which is a lower-bound estimator and considered one of the most reliable and less affected by bias among all nonparametric indices (27, 64–66). The values of the estimators from the samples to plot the curves shown in Fig. 2 were randomized, interpolated, and extrapolated with the package iNext in R (67).

The Chao2 estimator (bias corrected) is calculated by the following formula:where S are the actual numbers of species observed in the samples (m) and Q1 and Q2 are the species that appear in only one (unique) and two (duplicate) sampling units, respectively (27, 29). The 95% CI (CI bias corrected) of this index can be calculated by the formula

This estimation yielded a global value of 89,147 ± 1,101.5 species (Chao2 ± 95% CI) (Table 1). We are well aware that some studies provide different preferred estimators (68–70). However, many analyses, including simulation-based experiments, encourage the use of Chao2 to minimize bias (a summary is in ref. 71). This is the reason we considered the Chao2 index (based on occurrence data) our more useful estimator. Nonetheless, this estimator is sensitive to accurate quantification of the numbers of uniques and duplicates, and it is known that there are problems with false uniques in forest species richness datasets (24). Our Chao2 values may, thus, represent an overestimate to the degree that tree species recorded in only one sampling unit have been mistakenly identified as unique. Therefore, to check the reliability of our nonparametric estimates, we calculated the true number of uniques (28) in each continent and at a global scale to understand whether our values were influenced by the number of “falsely unique species.” This estimation of the true number of uniques is calculated with the formula adapted from ref. 28 for incidence-based data:where is the estimated true number of uniques; T is the number of sampling units (map cells); and Q2, Q3, and Q4 are observed duplicates, triplicates, and quadruplicates.

At the global level, the estimate of the true number of uniques is 13,162 compared with the observed 24,768. At the continental level, the number of estimated uniques was much lower than the observed one in South America (4,888 vs. 13,110) and somewhat lower in Eurasia (3,424 vs. 5,806), Africa (2,192 vs. 3,466), and Oceania (1,444 vs. 2,208), but it was slightly higher in North America (2,460 vs. 2,360). We then used the adjusted number of uniques in the Chao2 equation (see above) to calculate the Chao2adj estimates, (27–29).

We also calculated tree species rarity at continental and global scales for abundance (abundance-based rarity; i.e., based on the number of adjusted singletons [S1] and doubletons [S2]) and occurrence (occurrence-based rarity; i.e., based on the adjusted number of unique species and the number of duplicate ones). We defined the number of rare species as the sum of adjusted singletons and doubletons. We also computed an index of rarity importance using our occurrence-based dataset as the proportion of rare species over total richness and an S1 ratio, which is the proportion of singletons over doubletons.

Continental Biodiversity Partitioning.

We estimated the number of species shared among continents and unique to each continent using the Chao2 estimator (Fig. 4) from the occurrence-based data, and we represented them in a Venn diagram. We combined the observations of species richness for the five continents (n = 5) in all possible 25 − 1 = 31 combinations.

First, we calculated asymptotic species richness (Chao2) from occurrences observed in each continent; then, we intersected (creating a unique presence/absence binary entry for each species) the observed occurrences per each pair, triplet, quadruplet, and all five of continents (obtaining the occurrences of all the observed species in each combination of continents) and calculated the asymptotic species richness (Chao2) per each pair, triplet, quadruplet, and quintuplet continents. Therefore, a total of 31 estimates were obtained by the Chao2 index and plotted in a Venn diagram with the R package VennDiagram (72).

Additional cross-checks of the data pooling approach and q1/q2 relationship are in .