Thread by @hmlfpereira

Are most species rare? How many species are out there? Our paper on "Unveiling global species abundance distributions " is out today in #openaccess @NatureEcoEvo. A thread🧵 1/# @idiv @UniHalle @callaghanct

First a link to our paper 2/#nature.com/articles/s4155…

It's age-old question whether most species are rare or have some intermediate abundance (still relatively rare but not very rare). The log-series distribution proposed by Fisher says the former, while the log-normal distribution proposed by Preston says the latter. 3/#

In addition, in his classic 1948 paper in Ecology, Preston argued that when one had sampled only a small portion of the individuals, there would be a veil line that made the log-normal look like a log-series distribution. 4/#

It has been shown that if one assumes Poisson sampling, the log-series is a particular case of sampling a Gamma distribution (when k->0), with all other cases resulting in the negative binomial distribution. The sampling of a log-normal results in the Poisson log-normal.

Local studies showed these three distributions can fit the data, but little was known about the global scale (exception: birds and plants). Whether a veil exists has also been questioned. Using over 1 billion records in #gbif we tackled these questions across the tree of life.

We found that if we use only the first decades of records in @GBIF the distributions for many of the taxonomic groups look like log-series (monotonically decreasing), but as more data becomes available in later decades, they become uni-modal, and best explained by log-normal.

For instance, for amphibians and bony fishes if one uses observations only from the first couple decades of 20th century the distribution looks like a log-series, but if one looks at the records from last couple decades (when there is much more data) they show a unimodal shape.

This is a beautiful demonstration of the existence of a sampling veil, as predicted by Preston: the more one samples biodiversity, the more the veil moves to the left uncovering the true shape of the species abundance distribution (SAD). 9/#

For some of the groups, such as arachnids and insects, we have yet to unveil the unimodal distribution, and there most species still seem to be very rare as the number of species declines monotonically with their abundance.

So in the earlier decades of the 20th century, a goodness of fit test does not really separate which of the three distributions best explain the data, but as more and more data becomes available, the log-normal is a winner across the board.

Why are some groups unveiled and others not? Well it all depends on how many species there are in the group and how much effort have scientists put in studying that taxonomic group. 12/#

For instance for birds, we have pretty much discovered all species as they are not so many species and there are a lot of birders. In reality we observe the entire SAD every single year at this moment: the number of records of each species is proportional to their abundance❗️13/#

Or better seen in an animated GIF... 14/#

For more specious groups, such as insects, or poorly studied taxa, such as hornworts, or many of the non-vertebrate groups, we have yet to discover most species.

So can we estimate how many species are there on the planet yet to be discovered? We tried, but there are two major challenges we could not overcome. One is the difficult statistical behaviour of Poisson log-normal estimators.

But another one that we came to appreciate is that the concept of species has evolved over time, making it a moving target. We don't classify species in the same way today as we did 100 years ago, or even 20 years ago.

So, it is very hard to assess how many species we may yet discover as by the time they are discovered what we think is a species may be different! Anyway, our data suggests that for some species rich groups, we have yet to discover more than half of the species .... 18/#

Or at least the majority of the species in those groups have no recent records in GBIF (last 20 years). 19/#

So what else can we learn from our results? We speculate that the universality of the log-normal fit across taxa may imply that allopatric speciation is the most common mode of speciation. This remains yet to be tested, but would have profound implications. 20/20