Biased-corrected richness estimates for the Amazonian tree flora

Abstract

Amazonian forests are extraordinarily diverse, but the estimated species richness is very much debated. Here, we apply an ensemble of parametric estimators and a novel technique that includes conspecific spatial aggregation to an extended database of forest plots with up-to-date taxonomy. We show that the species abundance distribution of Amazonia is best approximated by a logseries with aggregated individuals, where aggregation increases with rarity. By averaging several methods to estimate total richness, we confirm that over 15,000 tree species are expected to occur in Amazonia. We also show that using ten times the number of plots would result in an increase to just ~50% of those 15,000 estimated species. To get a more complete sample of all tree species, rigorous field campaigns may be needed but the number of trees in Amazonia will remain an estimate for years to come.

Figure 1

Fit of SAD models to the species abundances from samples from the three ATDN data sets. Left column: observed frequencies of species in each abundance class (octaves, grey bars) and frequencies predicted by analytical Logseries (LS), truncated negative binomial (TNB) and Poisson-lognormal (PLN). Right column: rank-abundance plot in log-log scale of the abundances of species (gray) and the predicted abundances at each rank by the same three models.

Figure 2

Bias-corrected estimates of total species richness for each method and each data set. TNB, LS: upscaling from the fit of Truncated Negative Binomial and the analytical Logseries to abundances in the sample; LSE: linear extension of the distribution of estimated population sizes; ABC: approximated Bayesian Computation for estimated population sizes distribution. ABC and all bias corrections are derived from simulated samples with conspecific clumping from a logseries community. Bars depict bias-corrected 95% confidence intervals or similar (credible interval for ABC).

Figure 3

Extension of assumed logseries of estimated population sizes to predict the number of species in Amazonian forests. Grey dots in both panels are the estimated total population sizes of species recorded in the 2019 ATDN data set. The solid blue line in the main figure is the rank-abundance relationship predicted by a logseries with the average of estimates of number of species (15,874 spp). Dotted lines in the main panel delimit the rank-abundance for minimum and maximum of lower and upper limits 95% of the estimates. The lines in the inset panel are the mean and lower and upper bounds for the values of abundances estimates of the recorded species, also from averaging over all estimation methods.

Figure 4

Expected species-accumulation curves from simulated samples logseries with conspecific clumping based on the 2019 data set. The lines show simulated samples of a logseries with the mean (solid line) and lower-upper bounds (dotted lines) of the estimated number of species in the plot sample. Blue dots show the observed values for the three data sets.