Brocklehurst Et Al (2017) - Differences in Sps Frequency Distributions - Fossil B Diversity

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324032853

Accounting for differences in species frequency distributions when


calculating beta diversity in the fossil record

Article  in  Methods in Ecology and Evolution · March 2018


DOI: 10.1111/2041-210X.13007

CITATIONS READS

12 231

3 authors:

Neil Brocklehurst Michael O. Day


University of Oxford Natural History Museum, London
82 PUBLICATIONS   437 CITATIONS    46 PUBLICATIONS   460 CITATIONS   

SEE PROFILE SEE PROFILE

Jörg Fröbisch
Museum für Naturkunde - Leibniz Institute for Research on Evolution and Biodiver…
134 PUBLICATIONS   1,646 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The mid-late Permian transition and the Capitanian mass extinction among tetrapods View project

Permocarboniferous actinopterygian fishes from Brazil. View project

All content following this page was uploaded by Neil Brocklehurst on 09 May 2018.

The user has requested enhancement of the downloaded file.


|
Received: 28 November 2017    Accepted: 16 March 2018

DOI: 10.1111/2041-210X.13007

RESEARCH ARTICLE

Accounting for differences in species frequency distributions


when calculating beta diversity in the fossil record

Neil Brocklehurst1  | Michael O. Day2,3 | Jörg Fröbisch1,2,4

1
Leibniz-Institut für Evolutions- und
Biodiversitätsforschung, Museum für Abstract
Naturkunde, Berlin, Germany 1. Beta diversity is a measure of the taxonomic differentiation between habitats/
2
Evolutionary Studies Institute & School
localities within an assemblage, and is normally calculated as a set of pairwise
of Geosciences, University of the
Witwatersrand, Johannesburg, South Africa taxonomic “distances” between the localities.
3
Department of Earth Sciences, The Natural 2. Due to incomplete sampling, beta diversity estimates for fossil assemblages will
History Museum (NHMUK), London, UK
always be higher than the true value. However, the difference between the ob-
4
Institut für Biologie, Humboldt-
Universitätzu Berlin, Berlin, Germany served and true distances will vary greatly depending on differences in the shape
of the relative abundance distribution.
Correspondence
Neil Brocklehurst 3. Using simulations, it is shown that incomplete sampling of a homogenous fauna
Email: neil.brocklehurst@mfn-berlin.de with more even relative abundances of taxa produces higher beta diversity than

Funding information one with very few extremely common taxa.


Deutsche Forschungsgemeinschaft, Grant/ 4. A new procedure is proposed for calculating beta diversity in the fossil record,
Award Number: FR 2457/5-1; South
African Agency for Science and Technology whereby the observed distances are compared to the distance obtained by ran-
Advancement; Department of Science domly resampling a homogenous fauna with the same abundance distribution.
and Technology, Republic of South Africa;
Palaeontological Scientific Trust; DST/ This procedure is named Relative Abundance Corrected beta diversity. Simulations
NRF Centre of Excellence in Palaeoscience; indicate that this method outperforms raw beta diversity estimates, providing
National Research Foundation
beta diversity estimates closer to the true values even when the assumptions of
Handling Editor: Daniele Silvestro the method are violated.
5. This method is applied to an empirical dataset: Permian–Triassic tetrapods from
the Karoo Supergroup of South Africa. Increased provincialism is found across the
Permian–Triassic boundary in contrast to uncorrected beta diversity patterns.

KEYWORDS
abundance, beta diversity, Karoo, provinciality, sampling, tetrapod

1 |  I NTRO D U C TI O N represents no taxa shared between the localities (for a summary of
the various distance metrics, see Koleff, Gaston, & Lennon, 2003).
Since the seminal paper of Whittaker (1960), species richness has When studying macroevolutionary patterns, particularly in the fos-
been discussed in terms of alpha, beta and gamma diversity. Gamma sil record, one must account for the incomplete sampling of taxa, and
diversity represents the total diversity of an assemblage, which is de- beta diversity is no exception. Even if two localities hosted an identi-
termined by the richness within each locality or habitat (alpha diver- cal set of taxa, the fact that numerous taxa will be missing from both
sity) and the differentiation between these localities (beta diversity). makes it probable that the true taxonomic distance of 0 will not be
Beta diversity is normally calculated as a set of pairwise taxonomic obtained (Chao, Chazdon, Colwell, & Shen, 2005; Lande, 1996; Wolda,
“distances” between the localities ranging between 0 and 1, where a 1981). The worse the sampling, the fewer taxa are likely to be sampled
distance of 0 indicates an identical set of taxa in two localities and 1 in both localities and the higher the observed distance will be. Thus,

Methods Ecol Evol. 2018;1–12. © 2018 The Authors. Methods in Ecology and |  1
wileyonlinelibrary.com/journal/mee3  
Evolution © 2018 British Ecological Society
| Methods in Ecology and Evolu on
2       BROCKLEHURST et al.

when comparing the beta diversity of two assemblages/time bins, one allowing one to identify issues of sampling and evenness relatively
must account for the fact that the assemblages (and localities within) unhampered by confounding factors such as variations in preserva-
may not have been sampled equally (Chao et al., 2005). tion potential (see “Discussion”).
However, sampling standardization does not mitigate another A comprehensive database of specimens from the museum col-
issue: the impact of incomplete sampling will vary depending on differ- lections of South Africa was used, modified from Day et al. (2015),
ences in the relative abundances of the taxa in the assemblage. Since which indicates the taxonomic affinity of each specimen, its biostra-
beta diversity is a taxonomic comparison of assemblages, the identi- tigraphic assemblage zone (AZ; used as time bins) and the farm from
ties of the taxa drawn during subsampling will affect the result. All else which it was collected (these represent the localities). The taxonomy
being equal, an extremely common taxon should have a higher prob- was updated based on the published literature and direct examina-
ability of making it into the fossil record and being sampled (Kidwell, tion of the specimens. Specimens of uncertain affinity or stratigra-
2001, 2002). Therefore, when comparing localities that hosted an phy were discarded, as were single-­t axon localities. The datasets for
identical set of taxa, if a small number of these are highly abundant each assemblage-­zone are presented in Data S1–S8.
there is a greater probability of most of these making it into the fos-
sil record at each locality sampled, thus keeping the inferred beta di-
2.2 | Assessing the impact of changes in the
versity low. However, if there is a more even abundance distribution,
abundance distribution
there is a higher probability that each locality would preserve a differ-
ent set of taxa and inferred beta diversity would be higher. In short, The Karoo dataset was used as the basis for simulations designed to
beta diversity values calculated for two different fossil assemblages test the impact of changes in the relative abundances of taxa on in-
with different abundance patterns will not be directly comparable, ferences of beta diversity. For each time bin, a dataset was simulated
even if sampling standardization is employed, as the null expectation with the same sample size as the original data from that bin, but the
(the expected beta diversity value at a particular level of sampling in a specimens present in each locality were drawn at random from the
homogenous set of localities) will be different each time. original frequency distribution, i.e. the probability of drawing a spec-
The effect of variation in abundance distributions on inferences imen of a particular taxon is that taxon’s frequency in the original
of alpha and gamma diversity in the fossil record has long been ac- dataset. Thus, the fauna was assumed to be homogenous, and any
knowledged and discussed (Alroy, 2000; Alroy et al., 2001; Bush, differences between the localities in the time bin would simply be
Markey, & Marshall, 2004), and methods more robust to its influ- due to the incomplete sampling. Having created a simulated dataset,
ence have been suggested (e.g., Alroy, 2010; Chao & Jost, 2012). four distance metrics were used to infer the pairwise taxonomic dis-
However, its effect has received limited discussion in relation to tance between each locality, the mean of the pairwise distances rep-
beta diversity. Although Beck, Holloway, and Schwanghart (2013) resenting the beta diversity: the Sørenson index (Sørenson, 1948);
demonstrated reduced accuracy of beta diversity estimates when the Lennon index (Lennon, Koleff, Greenwood, & Gaston, 2001); a
more even abundance distributions are undersampled, but they did recent modification of the Forbes index, hereafter the Forbes* index
not suggest a solution. This paper demonstrates the influence of (Alroy, 2015); and the Bray–Curtis index (Bray & Curtis, 1957). The
variations in relative abundances of taxa on a variety of taxonomic first three of these are based on presence–absence data, while the
distance measures using both simulated data and empirical data Bray–Curtis index uses the observed abundances. The Forbes* and
from the Karoo Supergroup of South Africa. A method of correcting Sørenson indices are similarity measures, so to convert them to a
for this influence is described and tested via simulations. distance measure the values were subtracted from 1. For each time
bin, 100 of the above simulations were run.
Another set of simulations was run using an identical procedure
2 |  M ATE R I A L S A N D M E TH O DS
to that described above, but incorporating subsampling, in order to
ascertain whether differences in beta diversity between the time
2.1 | Dataset
bins were due to differences in sampling heterogeneity or the abun-
The empirical palaeontological data is derived from the Karoo dance distributions. As suggested by Alroy (2010) and Chao and Jost
Supergroup, a South African basin with a record covering the Middle (2012), the samples were standardized to fixed coverage rather than
Permian until the Middle Triassic. Within the Karoo sequence, the a fixed sample size using the algorithm proposed by Alroy (2014). For
Beaufort Group represents by far the best-­sampled record of ter- each locality within the dataset, individuals are drawn one at a time
restrial vertebrates from the Permian–Triassic, and has a well-­ until the sample is complete, each time calculating the coverage of
established biostratigraphy recently correlated with radiometric the sample using Good’s u (the proportion of singletons relative to
dates (Day et al., 2015; Rubidge, Erwin, Ramezani, Bowring, & de the total number of individuals, Good, 1953). Beta diversity is cal-
Klerk, 2013). Apart from the extensively studied taxonomy and spe- culated at each point where u crosses the specified coverage. The
cies richness patterns (e.g. Day et al., 2015; Fröbisch, 2013, 2014; data were standardized to a coverage of 0.6 (the highest coverage
Irmis, Whiteside, & Kammerer, 2013), the Karoo is a particularly allowing inclusion of all AZs), with 10,000 subsampling trials.
good subject for examination of the issues under study as it is a sin- To test how changes in the relative abundance distribution af-
gle basin with a consistent mode of preservation throughout, thus fect the result, the mean of the values obtained by the simulations
BROCKLEHURST et al. Methods in Ecology and Evolu on |
      3

was compared to the Evar evenness index each AZ. A fauna with a designed to illustrate the ability of RAC beta diversity to correctly
high value of Evar index contains a set of species with very similar identify when assemblages exhibit no difference in provinciality, and
relative abundances (Smith & Wilson, 1996). The correlation of mean the second to illustrate the ability of RAC beta diversity to correctly
simulation values for each AZ with the respective Evar values was identify the presence and extent of differences.
assessed using the Pearson product moment correlation coefficient, Under the first class, an assemblage containing up to 1,000,000
calculated in r version 3.3.2 (R Core Team, 2016). individuals in 50 species spread across two localities is generated.
The evenness of the relative abundance distribution is controlled by
drawing numbers of individuals of each species from a log-­normal
2.3 | A new procedure for calculating beta diversity
distribution, the variance of which represents an inverse measure of
We propose that when calculating the beta diversity of an assem- the evenness. Species are assigned to localities, with the proportion
blage, one should not simply use the raw mean pairwise distance of species shared between the two localities being a controlled pa-
value calculated; instead, one should ascertain how much higher this rameter. A fixed proportion of the individuals are sampled from each
raw value is than that which would be expected were a homogenous locality, representing the incompleteness of sampling. The raw beta
assemblage with the same relative abundance distribution to be diversity and the RAC beta diversity between the two localities is
sampled to the same extent as the empirical data. The simulations calculated, (both using Forbes* distances).
described above provide an excellent framework for this calculation. In such simulations where the proportion of shared species and
The following procedure is therefore proposed for or each the individual sampling probabilities are identical, the beta diversity
­assemblage/time bin: estimates of all should be identical. In practice, the raw beta diver-
sity values are expected to decrease with decreasing evenness (i.e.
1. Calculate the raw beta diversity using one of the available higher variances of the log-­normal distribution). Thus, a regression
distance metrics. line between the variance and the raw beta diversity estimates ob-
2. For each locality within the assemblage draw the same number of tained is expected to have a negative slope. For the RAC beta di-
specimens present at random from a frequency distribution rep- versity to be considered to outperform the raw beta diversity in
resenting the observed relative abundances of the taxa. the context of these simulations (correctly identifying when assem-
3. Calculate beta diversity of the simulated dataset. blages have identical beta diversities), the slope the regression line
4. Repeat steps two and three 100 times or more, and calculate the between the variance and the RAC beta diversity estimates should
mean of the results. This value will represent the null expecta- be reduced (ideally 0).
tion: the beta diversity to be expected if the fauna was homoge- One potential concern might be that the RAC method assumes
nous but sampled as incompletely as the observed data. 95% that the observed species frequency distribution is representative
confidence intervals around the mean are also calculated. of the true species frequency distribution i.e. that the probability of
5. Subtract the values obtained in step four from that in step one. fossilisation and sampling of each taxon is identical and therefore
Since at this point the maximum possible beta diversity value will the probability of a species being sampled is entirely dependent on
equal one minus the result of step four, and will therefore be dif- its abundance. The impact of inaccurate abundance distributions
ferent for each dataset, the values are rescaled using the equa- was tested by assigning each taxon a probability of individual pres-
tion: (O − N)/(1 − N), where O is the observed beta diversity, and ervation drawn from a normal distribution. Simulations were also run
N is the result of step 4. allowing the preservation probability of each taxon to differ depend-
ing on which locality was being sampled by drawing the preservation
The procedure produces a beta diversity metric where the probabilities separately; thus, the relative abundance distributions
maximum value (obtained when no overlap is observed between observed will potentially not only differ from the true distribution
species) is 1 and a homogenous fauna will have confidence inter- but also differ from each other.
vals bracketing 0. This entire procedure has been automated in r To illustrate the accuracy of the reconstructions of raw and RAC
using functions from the package vegan (Oksanen et al., 2017). The beta diversity estimates at various levels of coverage (proportion of
code for the function is provided in Data S9. The function allows species sampled), abundance distributions at fixed levels of even-
the option of carrying out standardization of the coverage of the ness were generated as described above, and all species assigned
localities (both in the empirical and simulated datasets). Hereafter to both localities (true beta diversity of zero). Specimens were then
this method is called Relative Abundance Corrected beta diversity drawn from the species pool one by one until fixed levels of coverage
(RAC beta diversity). were reached, at which point raw and RAC beta diversity estimates
were calculated.

2.4 | Testing the performance of RAC beta diversity


in simulations 2.5 | Beta diversity of the Karoo
Two classes of simulation were carried out to test the performance The RAC beta diversity through time was calculated for the Karoo
of RAC beta diversity relative to raw beta diversity. The first class is Supergroup to examine the changes in faunal provinciality through
| Methods in Ecology and Evolu on
4       BROCKLEHURST et al.

F I G U R E   1   Mean beta diversity values


of the 100 simulated datasets drawn
from a homogenous fauna with the
relative abundance distributions of each
assemblage zones (AZ) from the Karoo
Supergroup. (a) Raw; (b) subsampled. Beta
diversity estimate represents the mean
pairwise distance using four distance
metrics. Eod, Eodicynodon AZ; Tap,
Tapinocephalus AZ; Prist, Pristerognathus
AZ; Trop, Tropidostoma AZ; Cist,
Cistecephalus AZ; Dapt, Daptocephalus AZ;
Lyst, Lystrosaurus AZ; Cyn, Cynognathus
AZ

time. Beta diversity was inferred using both the raw data and data
3.2 | Assessing the performance of the RAC beta
with coverage standardized to 0.6 (the highest coverage allowing
diversity method
inclusion of all AZs), with 10,000 subsampling trials.
Under all simulations, the RAC beta diversity method was shown
to outperform the raw beta diversity estimate. The first class of
3 |   R E S U LT S
simulations indicate that it is better to be able to infer a lack of
difference in beta diversity in assemblages of different evenness
3.1 | Assessing the impact of evenness on inferred
(Table 2). When sampling is 10%, in fact, the regression line com-
beta diversity
paring evenness to the RAC beta diversity estimate is almost flat
The simulated beta diversity null expectation calculated for each (Figure 3a,b), even when the preservation potential of the taxa
Karoo AZ, despite the simulations assuming a homogeneous fauna varies (Figure 3e), indicating the method is consistently identify-
in each, show considerable variation (Figure 1). Although the ab- ing that assemblages with different evenness have the same beta
solute numbers differ, each distance measure produces the same diversity. At poorer levels of sampling (Figure 3c,d,f,h), the slope
pattern: the highest values in the Tapinocephalus and Daptocephalus of the regression line increases, indicating that resilience of the
AZs and the lowest values in the Eodicynodon, Pristerognathus and method to variations in the relative abundance distribution is de-
Tropidostoma AZs. This is not changed by subsampling. creasing, but in no set of parameters tested does the slope exceed
The simulated beta diversity values, both with and without that of the raw estimate. However, even though the method is
subsampling, exhibit an extremely strong, highly significant cor- robust to different taxa having different preservation potentials,
relation with the Evar index (Table 1, Figure 2), indicating that the it is less robust to having differing preservation potential between
variation in mean pairwise distances is strongly influenced by the the two localities (Figure 3g,h), albeit still outperforming raw beta
structure of the relative abundance distribution, and the issue is diversity.
not solved by coverage standardization. The correlation is positive, The second class of simulations demonstrates that the RAC
supporting the predictions made in the Introduction: a more even method produces results closer to the true beta diversity at all levels
relative abundance distribution results in higher beta diversity of coverage. When evenness is high, raw beta diversity estimates
estimates. show an almost linear relationship with coverage (Figure 4a). As
BROCKLEHURST et al. Methods in Ecology and Evolu on |
      5

TA B L E   1   Results of correlation tests between mean pairwise


distances between localities in homogenous faunas simulated from
the Karoo data and the evenness of the assemblages

Correlation test Pearson’s R p value

Forbes* distance (raw) ~ Evenness .837 .010


Forbes* distance (subsampled) .706 .050
~ Evenness
Sørenson distance (raw) ~ Evenness .868 .004
Sørenson distance (subsampled) .798 .031
~ Evenness
Lennon distance (raw) ~ Evenness .823 .004
Lennon distance (subsampled) .798 .031
~ Evenness
Bray–Curtis distance (raw) .836 .010
~ Evenness
Bray–Curtis distance (subsampled) .731 .049
~ Evenness
RAC Forbes* beta diversity (raw) .080 .850
~ Evenness
RAC Forbes* beta diversity .002 .996
(subsampled) ~ Evenness

RAC, Relative Abundance Corrected.

evenness decreases, the beta diversity inferred at low levels of cov- F I G U R E   2   Comparison of the mean beta diversity values (mean
erage is closer to the true value of zero, but the rate of improvement pairwise taxonomic distance) obtained from the homogenous
datasets simulated from the Karoo data using the Forbes* distance
as coverage increases is not so rapid (Figure 4b–d). This supports the
metric, and the evenness of each assemblage measured. (a) Raw;
results of Beck et al. (2013), who suggested that it was at low lev- (b) subsampled. Dashed lines represent regression lines.
els of evenness that beta diversity estimates were most inaccurate Abbreviations as in Figure 1
when sampling is poor. RAC beta diversity gives similar results at all
levels of evenness; at all coverage levels, RAC beta diversity is closer
to the true value of zero than raw beta diversity, and the variation in curves indicate a rise in RAC beta diversity across the Permian/
the values obtained is lower (Figure 4). The rate of improvement as Triassic boundary.
coverage increases is not so fast as that observed in the raw diversity
estimate until a coverage of about 0.9 is reached, at which point the
4 | D I S CU S S I O N
curves show an inflection.

4.1 | Abundance and beta diversity


3.3 | Empirical analyses
The study of variations in faunal provinciality through geological
The empirical beta diversity estimates for the Karoo, calculated as time is of great significance in studying macroevolutionary patterns,
the mean pairwise Forbes* distances between localities (Figure 5; with some studies suggesting that beta diversity may have had
see Data S10 for results from other distance metrics), both with more influence than alpha diversity on changes in gamma diversity
and without subsampling, show similar trends to the curve pro- (Bambach, 1977; Sepkoski, 1988). Several studies have highlighted
duced by simulating homogenous faunas. This indicates that a substantial changes in beta diversity coinciding with major radia-
large amount of the variation observed when raw beta diversity tions and mass extinctions (e.g., Button, Lloyd, Ezcurra, & Butler,
is calculated is a result of variations in the relative abundance 2017; Clapham, Shen, & Bottjer, 2009; Na & Kiessling, 2015; Sahney
distribution rather than variation in provinciality. No significant & Benton, 2008; Sahney, Benton, & Falcon-­L ang, 2010; Sidor et al.,
correlation is found between RAC beta diversity and evenness 2013). When considering this, the impact of changes in relative
(Table 1). abundance distributions on estimates of beta diversity in the fos-
Both raw and subsampled data indicates a fall in RAC beta sil record is particularly concerning. It is during such environmental
diversity between the Eodicynodon and Pristerognathus AZs and faunal upheavals that significant changes in abundance distribu-
(Figure 6). The subsampled data indicates RAC beta diversity sta- tions occur (Preston, 1948; McKinney & Lockwood, 1999; McElwain
bilized at this point until the end of the Permian, while the raw data & Punyasena, 2007; Wilson, 1992) e.g. the prevalence of pioneer/
suggests the decrease continued until the Cistecephalus AZ. Both disaster taxa following environmental upheaval (Hallam & Wignall,
| Methods in Ecology and Evolu on
6       BROCKLEHURST et al.

TA B L E   2   Slope (m) of the regression line comparing beta diversity estimates of simulated faunas to using the Forbes* distance metric to
the variance of the log-­normal distribution used to simulate the abundance distribution (an inverse measure of evenness); ideally slope
should be 0

Sampling probability of Sampling probability Proportion of taxa shared m of raw beta m of RAC beta
individuals of each taxon varies by locality? between localities diversity ~ variance diversity ~ variance

10% NA 1 −0.022 −0.005


10% NA 0.5 −0.024 −0.005
5% NA 1 −0.061 −0.032
5% NA 0.5 −0.067 −0.040
Drawn from normal distribution No 1 −0.021 −0.005
(M = 0.1)
Drawn from normal distribution Yes 1 −0.034 −0.019
(M = 0.1)
Drawn from normal distribution No 1 −0.057 −0.028
(M = 0.05)
Drawn from normal distribution Yes 1 −0.083 −0.058
(M = 0.05)

RAC, Relative Abundance Corrected.

1997; Harries, Kauffman, & Hansen, 1996; Pielou, 1975; Rodland & distances, incorporating the absolute frequencies rather than
Bottjer, 2001; Schubert & Bottjer, 1995). presence–absence data.
When simulating homogenous faunas from Karoo data, the The fact that incomplete sampling means that a homogenous as-
lowest mean pairwise distances are produced when drawing spec- semblage will not produce the true beta diversity value of 0 is not a
imens from Eodicynodon, Pristerognathus and Tropidostoma AZ fau- new observation (Chao et al., 2005). Nevertheless, all studies of beta
nas. These AZs also have the lowest evenness, being characterized diversity still make comparisons where 0 is treated as the null ex-
by single overwhelmingly abundant taxa. In the Pristerognathus and pectation. RAC beta diversity accounts for the fact that the null ex-
Tropidostoma AZs, the taxon is Diictodon feliceps, of which more than pectation is unique to each assemblage, and the faunal provinciality
200 specimens are known from each; the second most abundant taxa of an assemblage should be measured as deviation of the observed
in the two AZs have produced 39 and 54 specimens respectively. On beta diversity value from the null expectation of that specific assem-
the other hand, high mean pairwise distances are produced when blage. The simulations show that under all sampling regimes tested,
drawing from the Tapinocephalus and Daptocephalus AZ faunas, the RAC beta diversity estimate show a weaker relationship with the
where no such dominance is present. In the former Diictodon is still evenness of the relative abundance distribution than the raw esti-
the most abundant taxon, but there is not such a gap between it mate (although the strength of the relationship does increase when
and the other species, while in the latter the dicynodonts Oudenodon sampling levels are lowered). Moreover, at all levels of coverage, the
bainii, Dicynodon lacerticeps and Diictodon feliceps assemblages with RAC beta diversity estimate produces more accurate results than
two localities have very similar abundances. The simulations based the raw estimates. It should, however, be noted that at lower cover-
on assemblages of two localities support these results; under all age levels (below 0.9), the simulations suggest that the RAC method
sampling regimes tested, raw beta diversity will vary heavily accord- will still overestimate beta diversity, albeit not to the same degree as
ing to the evenness of the relative abundance even when the true a raw estimate.
provinciality does not. The RAC beta method starts with certain assumptions, perhaps
These results bear out the predictions made in the the most important of which is that the abundance distribution pre-
Introduction: when homogenous assemblages with more even served in the fossil record is the same or similar shape to the original
abundance distributions are subjected to incomplete sampling, distribution. This is an issue that has been discussed extensively in
beta diversity estimates will be higher than similarly sampled less the published literature (e.g. Bambach & Bennington, 1996; Kidwell,
even assemblages; thus, the two are not directly comparable. In 2001, 2002; McKinney, 1997; McKinney, Lindgard, Sepkoski, &
fact, the beta diversity curves simulated from Karoo data but Taylor, 1998; Vermeij & Herbert, 2004). While it does seem reason-
assuming homogenous faunas in each time bin, show very simi- able that more abundant species will have a higher probability of
lar trends to the actual beta diversity estimates (Figure 5). This making it into the fossil record, thus preserving the distribution, po-
would indicate that much of the observed variation in faunal pro- tential confounding issues including variable preservation potential
vinciality is actually a result of the shift in null expectation rather of different biogenic material in different environments (e.g. Cherns
than variation in provinciality. What is particularly surprising is & Wright, 2000, 2009; Wright, Cherns, & Hodges, 2003), differ-
that the same pattern is observed when using the Bray–Curtis ent preservation potential of different body sizes (Brocklehurst,
BROCKLEHURST et al. Methods in Ecology and Evolu on |
      7

F I G U R E   3   Raw and Relative Abundance Corrected (RAC) beta diversity estimates of simulated data compared to evenness of the fauna.
Coloured points represent the beta diversity estimates of individual simulations, and the lines represent the regression line between beta
diversity and evenness. The true proportion of taxa shared between the two localities is 100% (a,c,e,f,g,h) and 50% (b,d). a, b, c and d have
fixed sampling probabilities of 1% (a,b) and 0.5% (c,d). In e–h, the sampling probabilities of each taxon are drawn from a normal distribution
with a mean of 0.01 (e,g) or 0.005 (f,h). In g and h, the sampling probabilities of each taxon differs by locality
| Methods in Ecology and Evolu on
8       BROCKLEHURST et al.

F I G U R E   4   Raw (red) and Relative Abundance Corrected (RAC) beta diversity estimates of simulated datasets, sampled to different
coverages. True beta diversity in each case is 0. Variance of log-­normal distribution used to determine evenness of abundance distribution
(high variance = low evenness): (a) 1; (b) 2; (c) 3; (d) 4; (e) 5

Upchurch, Mannion, & O’Connor, 2012; Brown, Evans, Campione, is a bias towards the preservation of larger specimens in the Karoo
O’Brien, & Eberth, 2013; Kammerer, 2016) and the human collecting (Kammerer, 2016), then variability in body size will affect the raw
biases e.g. preference for collection of impressive specimens for mu- beta diversity estimates in the same way as variability in abundance
seums (Brown et al., 2013). Nevertheless, several studies have exam- would. That is, one very large species would have the same effect as
ined ecology in the fossil record assuming the observed distributions one very abundant species: being more likely to be preserved in all
are accurate (e.g. Codron, Botha-­
Brink, Codron, Huttenlocker, & localities and thus lowering the beta diversity.
Angielczyk, 2016; Jernvall & Fortelius, 2002; Kidwell & Brenchley, It might be more accurate to state that the RAC beta diversity
1994, 1996). method is correcting for variable fossilisation potential of the spe-
Such assumptions have been justified by studies suggesting cies, although much of our discussion has assumed that this is due
that relative abundance distributions are preserved reasonably ac- to different relative abundances due to the studies cited above sug-
curately, even with the vagaries of the fossil record (Kidwell, 2001, gesting that this is a large factor in preservation potential. It should
2002). Another point worth mentioning is that the confounding fac- be noted that the assumptions would likely not hold in a study based
tors potentially impacting the abundance distribution are not nec- on a wider geographical spread, potentially containing multiple as-
essarily confounding the estimates of RAC beta diversity. If there semblages with varying modes of preservation. Nevertheless, the
BROCKLEHURST et al. Methods in Ecology and Evolu on |
      9

F I G U R E   5   Comparison of the beta


diversity calculated from the empirical
data from the Karoo dataset to the mean
values obtained from the homogenous
simulated datasets using the Forbes*
distance metric. Narrow lines represent
95% confidence intervals.
(a) raw; (b) subsampled (coverage = 0.6).
Abbreviations as in Figure 1

F I G U R E   6   Relative Abundance
Corrected (RAC) beta diversity through
time within the Karoo. The thick lines
represent the RAC beta diversity
estimate, the narrow lines represent 95%
confidence intervals. Abbreviations as in
Figure 1

simulations indicate that such confounding factors do not appear to and be less geographically restricted, Brown, 1984; Kotze, Niemelä,
render the RAC beta diversity estimate less accurate than the raw O’hara, & Turin, 2003), resource availability (taxa which use wide-
beta diversity estimate. spread and abundant resources will themselves be widespread and
One important consideration is the non-­independence between abundant, Hanski, Kpiki, & Halkka, 1993; Gaston, 1994). Thus, an
abundance and range size. It has long been observed that locally assemblage with a less even relative abundance distribution will
abundant species tend to have larger range sizes (Brown, 1984; genuinely have a lower beta diversity, as there are fewer taxa with
Gaston, 1994, 1996; Gaston, Blackburn, & Lawton, 1997; Lawton, exceptionally high abundance and thus fewer taxa with large range
1993). The reason for this relationship is not clear, with suggested sizes. This fact emphasises that variation in the abundance distri-
mechanisms including the resource breadth hypothesis (species with bution is not merely a “nuisance parameter” whose signal needs to
broader environmental tolerances will have higher local densities be removed, but is indeed an important ecological characteristic
| Methods in Ecology and Evolu on
10       BROCKLEHURST et al.

(Chao & Jost, 2015). This consideration does not invalidate the RAC This study was funded by Deutsche Forschungsgemeinschaft grant
method, since the method assesses the shift in null by assigning all number FR 2457/5-­1, awarded to J.F. Data collection was funded by
taxa to all localities rather than by treating abundant and rare taxa the DST/NRF Centre of Excellence in Palaeoscience, the National
differently or by rarefying the sample (which exaggerates the influ- Research Foundation (NRF) and the Palaeontological Scientific Trust
ence of common taxa, Alroy et al., 2001). Nevertheless, the issue (PAST) and its Scatterlings of Africa programmes.
does highlight the fact that one cannot summarize the complexity of
an ecosystem with a single number (Chao & Jost, 2015).
AU T H O R S ’ C O N T R I B U T I O N S

N.B., designed the methods, analyses and simulation, wrote the


4.2 | Faunal provinciality in the Karoo
manuscript; M.O.D., collected the data, wrote the manuscript; J.F.,
The Karoo Supergroup has produced an unparalleled sample of late collected the data, wrote the manuscript.
Palaeozoic–early Mesozoic tetrapods, and provides near-­continuous
sedimentation across the end-­Permian mass extinction (Rubidge,
DATA AC C E S S I B I L I T Y
1995; Rubidge, 2005). It is the primary source of data on macroevo-
lutionary patterns in terrestrial vertebrates during this crucial time, Data deposited in the Dryad Digital Repository (http://datadryad.
although recent studies have suggested that treating the Karoo as org/resource/doi:10.5061/dryad.gh0q91j) (Brocklehurst, Fröbisch,
representative of the global pattern is problematic (Brocklehurst, & Day, 2018).
Day, Rubidge, & Fröbisch, 2017; Sidor et al., 2013).
It has been suggested that tetrapod-­bearing formations across
Gondwana show an increase in faunal provinciality across the ORCID
Permian–Triassic boundary (Sidor et al., 2013). However, Button
Neil Brocklehurst  http://orcid.org/0000-0003-2650-5480
et al. (2017), employing a method accounting for the phylogenetic
non-­independence of the taxa in each locality, inferred increased
cosmopolitanism following the extinction. In these studies, the fau- REFERENCES
nas being compared were tetrapod-­bearing basins or bioregions, and
Alroy, J. (2000). New methods for quantifying macroevolutionary
so the Karoo was treated as a single entity. However, the fossilifer-
patterns and processes. Paleobiology, 26, 707–733. https://doi.
ous localities cover an area of more than 160,000 km2 and it is not org/10.1666/0094-8373(2000)026<0707:NMFQMP>2.0
unreasonable to expect separate faunal provinces within this basin, .CO;2
driven by differences in habitats or geographical barriers limiting Alroy, J. (2010). Geographical, environmental and intrinsic biotic con-
trols on Phanerozoic marine diversification. Palaeontology, 53,
dispersal.
1121–1235.
The RAC beta diversity method with subsampling applied Alroy, J. (2014). Accurate and precise estimates of origination and extinc-
indicates an increase in beta diversity of the Karoo across the tion rates. Paleobiology, 40, 374–397. https://doi.org/10.1666/13036
Permian/Triassic boundary. This contradicts the results obtained Alroy, J. (2015). A new twist on a very old binary similarity coefficient.
Ecology, 96, 575–586. https://doi.org/10.1890/14-0471.1
by Button et al. (2017), who suggested a more homogenous fauna
Alroy, J., Marshall, C. R., Bambach, R. K., Bezusko, K., Foote, M., Fürsich,
at this time. While it is possible that the differences are due to the F. T., … Jacobs, D. K. (2001). Effects of sampling standardization
different scale of the analyses (global vs. regional), another pos- on estimates of Phanerozoic marine diversification. Proceedings of
sibility is that the results of Button et al. (2017) were influenced the National Academy of Sciences of the United States of America, 98,
6261–6266. https://doi.org/10.1073/pnas.111144698
by the increased unevenness of the tetrapod faunas following the
Bambach, R. K. (1977). Species richness in marine benthic habitats
extinction and the prevalence of disaster taxa like Lystrosaurus through the Phanerozoic. Paleobiology, 3, 152–167. https://doi.
(Fröbisch, 2014; Wignall, 2007). Unfortunately, due to the lack of org/10.1017/S0094837300005236
abundance data in the Button et al. dataset, this cannot be tested. Bambach, R. K., & Bennington, J. B. (1996). Do communities evolve? A
major question in evolutionary paleoecology. In D. Jablonski, D. H.
It should be noted that, while the networking statistics employed
Erwin, & J. H. Lipps (Eds.), Evolutionary paleobiology (pp. 123–160).
by Button et al. (2017) are in theory subject to the same influ- Chicago, IL: University of Chicago Press.
ence of variations in the relative abundance distributions, their Beck, J., Holloway, J. D., & Schwanghart, W. (2013). Undersampling and
study also incorporated a correction for the phylogenetic non-­ the measurement of betadiversity. Methods in Ecology and Evolution,
independence of the taxa. Further study is required to examine 4, 370–382. https://doi.org/10.1111/2041-210x.12023
Bray, J. R., & Curtis, J. T. (1957). An ordination of the upland forest com-
how the phylogenetic networking method responds to the influ-
munities of southern Wisconsin. Ecological Monographs, 27, 325–349.
ence of the abundance distribution. https://doi.org/10.2307/1942268
Brocklehurst, N., Day, M. O., Rubidge, B. S., & Fröbisch, J. (2017). Olson’s
Extinction and the latitudinal biodiversity gradient of tetrapods
AC K N OW L E D G E M E N T S in the Permian. Proceedings of the Royal Society B, 284, 20170231.
https://doi.org/10.1098/rspb.2017.0231
We would like to thank Roger Benson, John Alroy, Roger Close and Brocklehurst, N., Fröbisch, J., & Day, M. O. (2018). Data from: Accounting
two anonymous reviewers for helpful comments and discussion. for differences in species frequency distributions when calculating
BROCKLEHURST et al. Methods in Ecology and Evolu on |
      11

beta diversity in the fossil record. Dryad Digital Repository, http://dx. Gaston, K. J. (1996). The multiple forms of the interspecific abundance-­
doi.org/10.5061/dryad.gh0q91j distribution relationship. Oikos, 75, 211–220. https://doi.
Brocklehurst, N., Upchurch, P., Mannion, P. D., & O’Connor, J. (2012). The org/10.2307/3546192
completeness of the fossil record of Mesozoic birds: Implications for Gaston, K. J., Blackburn, T. M., & Lawton, J. H. (1997). Interspecific
early avian evolution. PLoS ONE, 7, e39056. https://doi.org/10.1371/ abundance-­range size relationships: An appraisal of mechanisms.
journal.pone.0039056 Journal of Animal Ecology, 66, 579–601. https://doi.org/10.2307/5951
Brown, J. H. (1984). On the relationship between abundance and dis- Good, I. J. (1953). The population frequencies of species and the estima-
tribution of species. American Naturalist, 124, 255–279. https://doi. tion of population parameters. Biometrika, 40, 237–264. https://doi.
org/10.1086/284267 org/10.1093/biomet/40.3-4.237
Brown, C. M., Evans, D. C., Campione, N. E., O’Brien, L. J., & Eberth, D. Hallam, A., & Wignall, P. B. (1997). Mass extinctions and their aftermath.
A. (2013). Evidence for taphonomic size bias in the Dinosaur Park Oxford, UK: Oxford University Press.
Formation (Campanian, Alberta), a model Mesozoic terrestrial alluvial-­ Hanski, I., Kpiki, J., & Halkka, A. (1993). Three explainations of the posi-
paralic system. Palaeogeography, Palaeoclimatology, Palaeoecology, tive relationship between distribution and abundance of species. In
372, 108–122. https://doi.org/10.1016/j.palaeo.2012.06.027 R. Ricklefs & D. Schulter (Eds.), Historical and geographical determi-
Bush, A. M., Markey, M. J., & Marshall, C. R. (2004). Removing bias from nants of community diversity (pp. 108–116). Chicago, IL: University of
diversity curves: The effects of spatially organized biodiversity on Chicago Press.
sampling-­standardization. Paleobiology, 30, 666–686. https://doi.org Harries, P. J., Kauffman, E. G., & Hansen, T. A. (1996). Models for bi-
/10.1666/0094-8373(2004)030<0666:RBFDCT>2.0.CO;2 otic survival following mass extinction. Geological Society, London,
Button, D., Lloyd, G. T., Ezcurra, M. D., & Butler, R. J. (2017). Mass ex- Special Publications, 102, 41–60. https://doi.org/10.1144/GSL.
tinctions drove increased global faunal cosmopolitanism on the su- SP.1996.001.01.03
percontinent Pangaea. Nature Communications, 8, 733. https://doi. Irmis, R. B., Whiteside, J. H., & Kammerer, C. F. (2013). Non-­ biotic
org/10.1038/s41467-017-00827-7 controls of observed diversity in the paleontologic record: An
Chao, A., Chazdon, R. L., Colwell, R. K., & Shen, T. J. (2005). A new sta- example from the Permo-­Triassic Karoo Basin of South Africa.
tistical approach for assessing similarity of species composition with Palaeogeography, Palaeoclimatology, Palaeoecology, 372, 62–77.
incidence and abundance data. Ecology Letters, 8, 148–159. https://doi.org/10.1016/j.palaeo.2012.07.014
Chao, A., & Jost, L. (2012). Coverage-­based rarefaction and extrap- Jernvall, J., & Fortelius, M. (2002). Common mammals drive the evolu-
olation: Standardising samples by completeness rather than size. tionary increase of hypsodonty in the Neogene. Nature, 417, 538–
Ecology, 93, 2533–2547. https://doi.org/10.1890/11-1952.1 540. https://doi.org/10.1038/417538a
Chao, A., & Jost, L. (2015). Estimating diversity and entropy profiles via Kammerer, C. F. (2016). A new taxon of cynodont from the Tropidostoma
discovery rates of new species. Methods in Ecology and Evolution, 6, Assemblage Zone (upper Permian) of South Africa, and the early evo-
873–882. https://doi.org/10.1111/2041-210X.12349 lution of Cynodontia. Papers in Palaeontology, 2, 387–397. https://doi.
Cherns, L., & Wright, V. P. (2000). Missing molluscs as evidence of large-­ org/10.1002/spp2.1046
scale, early skeletal aragonite dissolution in a Silurian sea. Geology, 28, Kidwell, S. M. (2001). Preservation of species abundance in marine death
791–794. https://doi.org/10.1130/0091-7613(2000)28<791:MM assemblages. Science, 294, 1091–1094. https://doi.org/10.1126/
AEOL>2.0.CO;2 science.1064539
Cherns, L., & Wright, V. P. (2009). Quantifying the impacts of early diage- Kidwell, S. M. (2002). Time-­averaged molluscan death assemblages:
netic aragonite dissolution on the fossil record. Palaios, 24, 756–771. Palimpsests of richness, snapshots of abundance. Geology, 30,
https://doi.org/10.2110/palo.2008.p08-134r 803–806. https://doi.org/10.1130/0091-7613(2002)030<0803
Clapham, M. E., Shen, S., & Bottjer, D. J. (2009). The double mass extinc- :TAMDAP>2.0.CO;2
tion revisited: Reassessing the severity, selectivity, and causes of the Kidwell, S. M., & Brenchley, P. J. (1994). Patterns in biolclastic accumu-
end-­Guadalupian biotic crisis (Late Permian). Paleobiology, 35, 32–50. lation through the Phanerozoic: Changes in input or distruction?
https://doi.org/10.1666/08033.1 Geology, 22, 1139–1143. https://doi.org/10.1130/0091-7613(1994)0
Codron, J., Botha-Brink, J., Codron, D., Huttenlocker, A. K., & 22<1139:PIBATT>2.3.CO;2
Angielczyk, K. D. (2016). Predator-­p rey interactions amongst Kidwell, S. M., & Brenchley, P. J. (1996). Evolution of the fossil re-
Permo-­ Triassic terrestrial vertebrates as a deterministic factor cord: Thickness trends in marine skeletal accumulations and their
influencing faunal collapse and turnover. Journal of Evolutionry implications. In D. Jablonski, D. H. Erwin, & J. H. Lipps (Eds.),
Biology, 30, 40–54. Evolutionary paleobiology (pp. 299–336). Chicago, IL: University of
Day, M. O., Ramezani, J., Bowring, S. A., Sadler, P. M., Erwin, D. H., Abdala, Chicago Press.
F., & Rubidge, B. S. (2015). When and how did the terrestrial mid-­ Koleff, P., Gaston, K. J., & Lennon, J. J. (2003). Measuring beta diversity
Permian mass extinction occur? Evidence from the tetrapod record of for presence-­absence data. Journal of Animal Ecology, 72, 367–382.
the Karoo Basin, South Africa. Proceedings of the Royal Society B, 282, https://doi.org/10.1046/j.1365-2656.2003.00710.x
20150834. https://doi.org/10.1098/rspb.2015.0834 Kotze, D. J., Niemelä, J., O’hara, R. B., & Turin, H. (2003). Testing
Fröbisch, J. (2013). Vertebrate diversity across the end-­ Permian abundance‐range size relationships in European carabid beetles
mass extinction – Separating biological and geological signals. (Coleoptera, Carabidae). Ecography, 26, 553–566.
Palaeogeography, Palaeoclimatology, Palaeoecology, 372, 50–61. Lande, R. (1996). Statistics and partitioning of species diversity, and
https://doi.org/10.1016/j.palaeo.2012.10.036 similarity among multiple communities. Oikos, 76, 5–13. https://doi.
Fröbisch, J. (2014). Synapsid diversity and the rock record in the org/10.2307/3545743
Permian-Triassic Beaufort Group (Karoo Supergroup), South Lawton, J. H. (1993). Range, population abundance and conserva-
Africa. In C. F. Kammerer, K. D. Angielczyk, & J. Fröbisch tion. Trends in Ecology and Evolution, 8, 409–413. https://doi.
(Eds.), The early evolutionary history of the Synapsida (pp. org/10.1016/0169-5347(93)90043-O
305–320). Dordrecht, The Netherlands: Springer. https://doi. Lennon, J. J., Koleff, P., Greenwood, J. J. D., & Gaston, K. J. (2001). The
org/10.1007/978-94-007-6841-3 geographical structure of British bird distributions: Diversity, spatial
Gaston, K. J. (1994). Rarity. London, UK: Chapman & Hall. https://doi. turnover and scale. Journal of Animal Ecology, 70, 966–979. https://
org/10.1007/978-94-011-0701-3 doi.org/10.1046/j.0021-8790.2001.00563.x
|
Methods in Ecology and Evolu on
12       BROCKLEHURST et al.

McElwain, J. C., & Punyasena, S. W. (2007). Mass extinction events and Sepkoski, J. J. (1988). Alpha, beta, or gamma: Where does all the di-
the plant fossil record. Trends in Ecology & Evolution, 22, 548–557. versity go? Paleobiology, 14, 221–234. https://doi.org/10.1017/
https://doi.org/10.1016/j.tree.2007.09.003 S0094837300011969
McKinney, M. L. (1997). How do rare species avoid extinction? A pale- Sidor, C. A., Vilhena, D. A., Angielczyk, K. D., Huttenlocker, A. K., Nesbitt,
ontological view. In W. E. Kunin & K. J. Gaston (Eds.), The biology of S. J., Peecook, B. R., … Tsuji, L. A. (2013). Provincialization of terres-
rarity (pp. 110–129). Dordrecht, The Netherlands: Springer. https:// trial faunas following the end-­Permian mass extinction. Proceedings
doi.org/10.1007/978-94-011-5874-9 of the National Academy of Sciences of the United States of America,
McKinney, F. K., Lindgard, S., Sepkoski, J. J., & Taylor, P. D. (1998). 110, 8129–8133. https://doi.org/10.1073/pnas.1302323110
Decoupled temporal patterns of evolution and ecology in two post-­ Smith, B., & Wilson, J. B. (1996). A consumer’s guide to evenness indices.
Paleozoic clades. Science, 281, 807–809. https://doi.org/10.1126/ Oikos, 76, 70–82. https://doi.org/10.2307/3545749
science.281.5378.807 Sørenson, T. (1948). A method of establishing groups of equal amplitude
McKinney, M. L., & Lockwood, J. L. (1999). Biotic homogenization: A in plant sociology based on similarity of species content and its appli-
few winners replacing many losers in the next mass extinction. cation to analyses of the vegetation on Danish commons. Biologiske
Trends in Ecology & Evolution, 14, 450–453. https://doi.org/10.1016/ Skrifter, 5, 1–34.
S0169-5347(99)01679-1 Vermeij, G. J., & Herbert, G. S. (2004). Measuring relative abundance in
Na, L., & Kiessling, W. (2015). Diversity partitioning during the Cambrian fossil and living assemblages. Paleobiology, 30, 1–4. https://doi.org/1
radiation. Proceedings of the National Academy of Sciences of the 0.1666/0094-8373(2004)030<0001:MRAIFA>2.0.CO;2
United States of America, 112, 4702–4706. https://doi.org/10.1073/ Whittaker, R. H. (1960). Vegetation of the Siskiyou mountains, Oregon
pnas.1424985112 and California. Ecological Monographs, 30, 279–338. https://doi.
Oksanen, J., Blanchet, F. G., Friendly, M., Kindt, R., Legendre, P., McGlinn, org/10.2307/1943563
D., … Wagner, H. (2017). vegan: Community Ecology Package. R Wignall, P. B. (2007). The End-­ Permian mass extinction–how bad
package version 2.4-6. did it get? Geobiology, 5, 303–309. https://doi.org/10.1111/j.
Pielou, E. C. (1975). Ecology diversity. New York, NY: John Wiley and Sons. 1472-4669.2007.00130.x
Preston, F. W. (1948). The commonness, and rarity, of species. Ecology, Wilson, E. O. (1992). The diversity of life. Cambridge, MA: Belnap Press.
29, 254–283. https://doi.org/10.2307/1930989 Wolda, H. (1981). Similarity indices, sample size and diversity. Oecologia,
R Core Team. (2016). R: A language and environment for statistical comput- 50, 296–302. https://doi.org/10.1007/BF00344966
ing. Vienna, Austria: R Foundation for Statistical Computing. Wright, P., Cherns, L., & Hodges, P. (2003). Missing molluscs: Field testing
Rodland, D. L., & Bottjer, D. J. (2001). Biotic recovery from the end-­ taphonomic loss in the Mesozoic through early large-­scale aragonite
Permian mass extinction: Behavior of the inarticulate brachio- dissolution. Geology, 31, 211–214. https://doi.org/10.1130/0091-761
pod Lingula as a disaster taxon. Palaios, 16, 95–101. https://doi. 3(2003)031<0211:MMFTTL>2.0.CO;2
org/10.1669/0883-1351(2001)016<0095:BRFTEP>2.0.CO;2
Rubidge, B. S. (1995). Biostratigraphy of the Beaufort Group (Karoo Supergroup).
Pretoria, South Africa: South African Committee for Straticgraphy.
Rubidge, B. S., (2005). 27th Du Toit Memorial Lecture: Re-uniting lost con- S U P P O R T I N G I N FO R M AT I O N
tinents–fossil reptiles from the ancient Karoo and their wanderlust.
South African Journal of Geology, 108, 135–172. Additional Supporting Information may be found online in the
Rubidge, B. S., Erwin, D. H., Ramezani, J., Bowring, S. A., & de Klerk, W. J. ­supporting information tab for this article.
(2013). High-­precision temporal calibration of Late Permian vertebrate
biostratigraphy: U-­Pb zircon constraints from the Karoo Supergroup.
Geology, 41, 363–366. https://doi.org/10.1130/G33622.1
How to cite this article: Brocklehurst N, Day MO, Fröbisch J.
Sahney, S., & Benton, M. J. (2008). Recovery from the most profound
mass extinction of all time. Proceedings of the Royal Society B, 275, Accounting for differences in species frequency distributions
759–765. https://doi.org/10.1098/rspb.2007.1370 when calculating beta diversity in the fossil record. Methods
Sahney, S., Benton, M. J., & Falcon-Lang, H. J. (2010). Rainforest collapse Ecol Evol. 2018;00:1–12. https://doi.org/10.1111/2041-
triggered Carboniferous tetrapod diversification in Euramerica.
210X.13007
Geology, 38, 1079–1082. https://doi.org/10.1130/G31182.1
Schubert, J. K., & Bottjer, D. J. (1995). Aftermath of the Permian-­Triassic
mass extinction event: Paleoecology of Lower Triassic carbonates in
the western USA. Palaeogeography, Palaeoclimatology, Palaeoecology,
116, 1–39. https://doi.org/10.1016/0031-0182(94)00093-N

View publication stats

You might also like