Over-estimating extinction rates

19 05 2011

I meant to get this out yesterday, but was too hamstrung with other commitments. Now the media circus has beat me to the punch. Despite the lateness (in news-time) of my post, my familiarity with the analysis and the people involved gives me a unique insight, I believe.

So a couple of months ago, Fangliang He and I were talking about some new analysis he was working on where he was testing the assumption that back-casted species-area relationships (SAR) gave reasonable estimates of inferred extinction rates. Well, that paper has just been published in today’s issue of Nature  by Fangliang He and Stephen Hubbell entitled: Species–area relationships always overestimate extinction rates from habitat loss (see also the News & Views piece by Carsten Rahbek and Rob Colwell).

The paper has already stirred up something of a controversy before the ink has barely had time to dry. Predictably, noted conservation biologists like Stuart Pimm and Michael Rosenzweig have already jumped down Fangliang’s throat.

Extinction rates of modern biota in the current biodiversity crisis (Ehrlich & Pringle 2008) are wildly imprecise. Indeed, it has been proposed that extinction rates exceed the deep-time average background rate by 100- to 10000-fold (Lawton & May 2008; May et al. 1995; Pimm & Raven 2000), and no rigorously quantification of these rates globally has ever been accomplished (although there are several taxon- and region-specific estimates of localised extinction rates (Brook et al. 2003; Regan et al. 2001; Hambler et al. 2011; Shaw 2005).

Much of the information used to infer past extinction rate estimates is based on  the species-area relationship (e.g., Brook et al. 2003); this method estimates extinction rates by reversing the species-area accumulation curve, extrapolating backward to smaller areas to calculate expected species loss. The concept is relatively simple, even though the underlying mathematics might not be.

The basic species-area relationship is a power function of the form: S = cAz

Figure 1. Multiple species-area relationship (SAR) models applied demonstrating model uncertainty (Giam et al. 2011)

where S = the number of species in a broad habitat type of area A, and c and z are estimated constants. Not only are there are various modifications of the underlying model (e.g., the exponential, negative exponential and monod forms – Giam et al. 2011; Guilhamon et al. 2008) (Fig. 1), they can be combined them using both Bayesian and maximum-likelihood approaches to provide model-averaged predictions of species accumulation with habitat area (Giam et al. 2011). Such an approach ensures that model uncertainty is also incorporated into the models that are known to rank differently depending on region and taxon modelled (Giam et al. 2011).

The approach to estimate extinctions generally takes the basic form of the power model above and adjusts it based on a difference between an ‘original’ time t and some modified (i.e., lost habitat) time t+1:

St+1/St = c(At+1/At)z

However, in 2010, this approach had a major paradigm shift (Koh & Ghazoul 2010; Koh et al. 2010). The first problem in the above approach assumes that the land use between the habitat areas being modelled (i.e., the ‘matrix’) are completely inhospitable to any species considered; the second assumes that there are no deleterious effects of the matrix on the habitat fragments (i.e., ‘edge effects’). A modified approach therefore corrects simultaneously for taxon-specific responses to individual components of the landscape matrix (Fig. 2) by adjusting the power coefficient (z), and for the matrix ‘edge effects’ on the At+1 term:

St+1/St = c((At+1/- δ ∑Mβj)/At)γNpiσi

Figure 2. Matrix-calibrated species-area relationship model (Koh & Ghazoul 2010)

where bj  = the edge-affected area of the jth habitat fragment, M = the total number of fragments, δ  = the declining function of the edge effect as perceived by the taxon (e.g., δ = 0.5 if edge effect declines linearly with distance from habitat margin), γ = a constant that reflects the rate of change in species number per unit area on true island archipelagos, σ = an area-weighted average of taxon-specific responses to matrix components, p = the proportional area of the ith matrix component relative to the total area of the matrix, and N = the total number of matrix components (Koh & Ghazoul 2010; Koh et al. 2010).

Now He & Hubbell (2011) set to challenge all extinction rate estimates based on species-area relationships. The problem is rather obvious, but has failed to be appreciated because the power function  has no sampling theory relating it to species distributions.

The species-area relationship essentially estimates the area required to sample the first individual of a species, whereas the endemics-area relationship is its mirror image and the area required to sample the last individual of that species (Fig. 3). This essentially allows for the estimation of extinction based on the loss of habitat area. However, the mirror image assumption is false when a species’ individuals are clumped (i.e., not randomly distributed), which is nearly ubiquitous in nature (Condit et al. 2000). Thus, the area that must be added to find the first individual of a species in general is much smaller than the area that must be removed to eliminate the last individual of a species. Therefore, on average it takes a much greater loss of area to cause the extinction of a species than it takes to add the species on first encounter, and this result holds regardless of how well the power-law species-area function fits species-area data.

Figure 3. Range distribution of a species (blue area), and an arbitrary starting sample point, indicated by +. Regardless of starting location, a sampling frame of arbitrary shape (here circular) with an area of a size sufficient to contact the species for the first time is always less than the sample area needed to encompass the entire range of the species (He & Hubbell 2011)

He & Hubbell (2011) estimated in a few examples that species extinctions are over-estimated by about 75 %, but this is likely to be taxon- and region-specific.

Now, this is exciting in its own right, but the story isn’t finished. I think the next step is to attempt to combine model uncertainty, matrix calibration and the EAR bias. Instead of complaining that the He & Hubbell approach negates recent developments, it should be viewed instead as another way to improve existing models. Better predictions of land-use change over the coming century, and the state-of-the-art species-area models, should be able determine the degree of bias in extinction predictions.

Of course, there are potentially some issues that remain unresolved. In speaking with some colleagues about the He & Hubbell paper, it seems that it might only be implemented with associated distribution data. Where these are not available or not of high enough quality, then perhaps the approach is not useable.

Another concern arises from the calculation of z itself – if it is estimated from species reduction observations after land-use change rather than accumulation curves, it could better reflect extinctions in the backward calculation. Thus, the bias proposed by He & Hubbell might not be as great as they claim.

So, expect a battery of critiques and responses to this article. Exciting stuff, indeed.

References

ResearchBlogging.orgHe, F., & Hubbell, S. (2011). Species–area relationships always overestimate extinction rates from habitat loss Nature, 473 (7347), 368-371 DOI: 10.1038/nature09985


Actions

Information

6 responses

30 05 2016
Species-area & species-accumulation curves not the same | ConservationBytes.com

[…] species-area relationships is the effect of different sampling regimes, a problem I’ve blogged about before. Whether you are sampling once-contiguous forest of habitat patches in a ‘matrix’ of […]

Like

22 08 2013
Biogeography comes of age | ConservationBytes.com

[…] blogged about SAR issues before – what makes it so engaging and controversial is that SAR is the principal tool to […]

Like

22 08 2013
Biogeography comes of age | Wiley at INTECOL 2013

[…] blogged about SAR issues before – what makes it so engaging and controversial is that SAR is the principal tool to […]

Like

9 07 2011
Life, death and Linneaus « ConservationBytes.com

[…] yet put it all together, but most of the base code is written. If you’ve been following the minor dust-up in the ecological literature about SAR, you’ll understand why the first photo above is […]

Like

21 05 2011
stuart pimm

The Hubbell paper is a sham: it does not report extinction rates or the numbers of species that are threatened. Despite its posturing, it deals with a different issue. The paper is riddled with false statements. For instance:

The paper states: “Estimates of extinction rates based on (the species-area) method are almost always much higher than those actually observed.” It is unequivocally false. One reference used to support this (Pimm and Askins) uses a species-area relationship to predict 4.5 bird extinctions following deforestation in Eastern North America and then notices that four species went extinct and one is threatened.

There are dozens of other studies of many taxa around the world that find equally compelling agreements between predicted and observed extinctions. A small selection of them follows.

So what does the paper model — and why does it poorly address the issue of extinctions? Imagine destruction that wipes out 95% of the habitat in an area metaphorically “overnight”. How many species have disappeared “the following morning”? The paper tells you. It is not many, just those wholly restricted to the 95% (and absent from the 5% where they would survive). The important question is …
How many of additional species living lonely lives in their isolated patches (the 5%) would become extinct eventually because their population sizes are too small to be viable? A different species-area curve applies — the one for islands, which are isolated. It is a much larger number of extinctions, of course, and the one used in the studies mentioned above that find such compelling agreement between predicted against observed extinctions.
By all means, feel free to share this.
A response will be submitted to Nature shortly.
Stuart

Pimm, S. L. & Askins, R. A. Forest losses predict bird extinctions in eastern North America. Proc. Natl Acad. Sci. USA 92, 9343–9347 (1995).

Brooks, T. M. et al. Habitat loss and extinction in the hotspots of biodiversity. Conserv. Biol. 16, 909–923 (2002).

Grelle, C. E. de V., Fonseca, G. A. B., Fonseca, M. T. & Costa, L. P. The question of scale in threat analysis: a case study with Brazilian mammals. Animal Conserv. 2, 149–152 (1999).

Brooks, T. & Balmford, A. Atlantic forest extinctions. Nature 380, 115 (1996).

Cowlishaw, G. Predicting the pattern of decline of African primate diversity: an extinction debt from historical deforestation. Conserv. Biol. 13, 1183–1193 (1999).

Brook, B. W., Sodhi, N. S. & Ng, P. K. L. Catastrophic extinctions follow deforestation in Singapore. Nature 424, 420–423 (2001)

Brooks, T. M., S. L. Pimm, V. Kapos and C. Ravilious 1999. Threat from deforestation to montane and lowland birds and mammals in insular Southeast Asia. J. Anim. Ecol. 68: 1061-1078

Brooks, T. M., Pimm, S. L., & Oyugi, J. O. Time lag between deforestation
and bird extinction in tropical forest fragments. Conserv. Biol. 13, 1140-1150
(1999).

A full discussion of species area curves appears in

Rosenzweig, M.L. Species diversity in space and time. (Cambridge Univ.
Press, 1995)

Like

19 05 2011
CJAB

Related blog post: Extinction and its Discontents (New York Times DotEarth blog)

Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s