Species accumulation over space and time in European Plio-Holocene mammals

Page 1

RESEARCH ARTICLE

Species accumulation over space and time in EuropeanPlio-Holocene mammals

P. Raia • F. Carotenuto • C. Meloro • P. Piras • C. Barbera

Received: 20 October 2009 / Accepted: 10 May 2010 / Published online: 22 May 2010� Springer Science+Business Media B.V. 2010

Abstract The rate of increase in species number with sampled area is one issue of major

interest in ecology. Species number increases with sampled time as well, though this kind

of analysis is much rarer in literature. Species-area and species-time relationships have

been recently integrated in a single model, which allows studying how time and area

interact with each other in determining the cumulative increase in species richness. Here

we studied species-area, species-time, and species-time-area relationships in Plio-Holocene

large mammals of Western Eurasia, by using an extensive database including 184 species

distributed in 685 fossil sites. We found that the increase of species number with time is

much higher than with area. When sampling inequality of fossil localities in time and space

is accounted for, time and area interact with each other in a negative, though non-linear

fashion. The intense climatic changes that characterized the Plio-Holocene period appar-

ently affected both species-area and species-time relationships in large mammals, by

increasing the slope of the former during the Pliocene and middle Pleistocene, and of the

latter during younger, climatically harsher, late Pleistocene times. This study emphasizes

the importance of accounting for time and space in tracing paleodiversity curves.

Electronic supplementary material The online version of this article (doi:10.1007/s10682-010-9392-3)contains supplementary material, which is available to authorized users.

P. Raia (&) � F. Carotenuto � C. BarberaDipartimento di Scienze della Terra, Universita degli Studi di Napoli ‘Federico II’,L.go San Marcellino 10, 80138 Naples, Italye-mail: [email protected]

P. Raia � P. PirasCenter for Evolutionary Ecology, Largo San Leonardo Murialdo 1, 00146 Rome, Italy

C. MeloroHull York Medical School, The University of Hull, Loxley Building Cottingham Road,Hull HU6 7RX, UK

P. PirasDipartimento di Scienze Geologiche, Universita Roma Tre, Largo San Leonardo Murialdo 1,00146 Rome, Italy

123

Evol Ecol (2011) 25:171–188DOI 10.1007/s10682-010-9392-3

Page 2

Keywords Species-area relationship � Species-time relationship � Species-time-area

relationship � Ice age mammals

Introduction

The increase in species number over time and space are fundamental topics in ecology.

Currently, accounts of rates of species accumulation over space are much more common

than over time. In fact, the species–area relationship (SAR) is one of the most robust

ecological rules ever investigated (Rosenzweig 1995). Species accumulate over space at a

grossly predictable rate, so that the exponents of the species-area curve are rather similar

even across very different combinations of organisms and habitats. The variation in the

SAR slopes mostly depends on the biogeographical scale of investigation. Highest slopes

(some 0.9 and higher) are obtained by sampling at larger spatial scales, when they actually

pool biogeographical provinces with different biological histories. Slopes computed by

sampling species over islands in an archipelago (Rosenzweig 1995; Triantis et al. 2008),

are intermediate (some 0.35). The shallower slopes (some 0.15) are typically obtained by

sampling within a single biogeographical province. The strength of the SAR made it a

viable tool in topics as disparate as the estimation of extinction risk in fragmented habitats

(Pimm and Askins 1995; Kinzig and Harte 2000), latitudinal diversity gradients (Rosen-

zweig and Sandlin 1997), construction of paleodiversity curves (Barnosky et al. 2005) and

conservation biology (Myers et al. 2000; Fattorini 2007).

Both mechanicistic and biological explanations for SAR parameters (e.g. its slope, and

its shape in either logarithmic or non-trasformed data spaces) have been proposed as well

(May 1975; Shmida and Wilson 1985; Harte et al. 1999; Plotkin et al. 2000; Hubbell 2001;

Scheiner 2003; Turner and Tjørve 2005; Dengler 2009). Preston (1960) first suggested an

explanation for SAR based on species abundance distribution. Then, he advanced a similar

reasoning (and data) to model the accumulation of species over time: the species-time

relationship STR. STR describes the rate at which new species are sampled in a given area

over a given extent of time. It is characterized by three separate phases (Rosenzweig 1998;

Carey et al. 2007) occurring at three different temporal scales. An initial phase spanning

over short time intervals, when the accumulation of species depends on sampling increase,

so that it is a sampling artefact. Then, a second phase encompassing ecological processes

such as immigration and extinction, which mostly accrues to species temporal turnover

(ecological successions), and finally a third phase, influenced by long-term evolutionary

processes such as speciation, and extinction (Rosenzweig 1998; White 2004; Carey et al.

2007; Magurran 2007). The existence of the former two phases (sampling and ecological)

in the STR was borne out by White (2004, 2007; White et al. 2006) but the third phase (i.e.

‘‘evolutionary’’) is still under-investigated in literature, beside the fact that STR accounts

are rather rare overall (Carey et al. 2007). Rosenzweig (1998) made a first remarkable

review and comparison of STRs and SARs to date. His STRs were computed over time

intervals spanning 11 orders of magnitude (from days to the whole Phanerozoic).

Rosenzweig argued that STR exponents are similar to SAR’s at the smallest temporal

scale, because both relationships are driven by pure sampling artefacts which, he noted,

‘‘fails to astonish or matter much to a biologist’’. At the intermediate temporal scale (from

weeks to years) he noted STR slope departs seriously from SAR, with exponents ranging

from 0.37 to 0.90 (Rosenzweig 1998). However, many different slope values have been

reported since this work was published. For instance, McKinney and Frederick (1999)

172 Evol Ecol (2011) 25:171–188

123

Page 3

sampled fossil foraminifera over time intervals as long as 4 million years and got STR

slopes close to 0.30, which are consistently linear in log-log plots. They claimed that their

STRs are similar to SARs computed across islands. Hadly and Maurer (2001) produced

STRs for fossil mammals across 16 stratigraphic levels in Lamar Cave, in the Great Basin

of North America, spanning ca 3,000 years. They fitted the data with the power function

and found an exponent of 0.27. Adler and Lauenroth (2003) computed STRs over decades

in plant experimental plots sampled in North America, and got slopes ranging from 0.19 to

0.42, again with the power function selected for fitting the data. White et al. (2006)

gathered data for 984 extant community time-series and noted that most STR exponents are

comprised in between 0.23 and 0.39, although steeper slopes accrue to both birds and land

plants data. They also stated that when species richness is high STR slope is low. Con-

versely, they postulated intense climatic variability might increase temporal turnover on

the one hand and might decrease richness on the other, thereby increasing the STR slope.

STR and SAR are separate statistical models, yet the mutual influence of time and area

on species accumulation rate can be studied testing both variables together (Rosenzweig

1998). Adler and Lauenroth (2003) proposed a species-time-area relationship (STAR)

model, which includes area, time and their interaction term, and found that time has a

strong effect on species richness even at broad spatial scales. Adler et al. (2005) further

tested their ‘‘STAR model’’ against simpler models sharing the rationale of accounting

both for temporal and spatial variation together, but with no interaction among them

(e.g. Rosenzweig 1998). Adler and Lauenroth’s STAR takes the form:

log S ¼ log cþ z1 log Aþ w1 log T þ uðlog AÞðlog TÞ

where S is species richness, c is the intercept, A is the sampled area, T is the sampled time

span, z1 is the scaling exponent of the SAR at the unit time scale, w1 is the scaling exponent

of the STR at the unit spatial scale, and u is the fitted interaction parameter. This model

works better than others in predicting species richness across temporal and spatial scales

(Adler et al. 2005). The fitted interaction term, u, is always negative, meaning the effect of

increasing spatial scale on STR slope, and the effect of increasing temporal interval on

SAR slope, are both negative in a variety of organisms and across different sampling

scales.

Here, we fitted SAR, STR, and STAR models to cumulative species richness data in

Plio-Holocene fossil large mammals of Western Eurasia. Our aim is to depict how the

sampling of new species varies over time and space, and to understand the effect of long-

term climate change on species accumulation rates at a variety of temporal and spatial

scales. The chosen fossil record is ideal to our goal because: (1) it is intensively-studied, as

the species under scrutiny attracted interest from many investigators; (2) it is very well-

sampled, because the Plio-Holocene is the most recent past period, therefore there is plenty

of fossil remains and (3) it is the most interesting for during this time span climate swung

repeatedly to extreme conditions worldwide, a pattern superimposed over a net trend

toward increasingly lower temperatures (Zachos et al. 2001). Therefore, this period is ideal

to study the effect of climate change on species accumulation rates over time and space.

Recently, a number of studies have demonstrated the strong influence these climatic

changes had on the evolution of mammalian communities (Bobe and Behrensmeyer 2004;

Lister 2004; Barnosky et al. 2005; Raia et al. 2005; Barnosky and Kraatz 2007; Meloro

et al. 2008). Raia et al. (2005) found that the Plio-Pleistocene Italian paleocommunities

(PCOM sensu Raia et al. 2005) underwent stronger turnover rates as the climate became

cooler. Furthermore, Meloro et al. (2008) recorded a net increase in Italian PCOMs

Evol Ecol (2011) 25:171–188 173

123

Page 4

temporal turnover with the onset of Galerian (some 1 million years ago [Ma]), a peculiar

moment in the history of global climatic change for it coincides with the beginning of the

strongest glacial phases. Indeed, some 1 Ma (in coincidence with the Jaramillo Paleo-

magnetic Event) the duration of complete climatic cycles (due to variation of obliquity and

precession of the Earth’s axis) changed from 20 to 40 kiloyears (ky), to new, longer cycles

of some 100 ky (due to Earth’s orbital eccentricity variation). With the increase of cycle

duration the global temperatures reached extreme values and there were wider climatic

shifts, a trend which even intensified since 0.5 Ma (Zachos et al. 2001). These new cycles

determined stronger than ever changes in habitats that influenced Pleistocene mammal

turnovers in Eurasia (Lister 2004). Yet, the intensity of temporal turnover is deemed to be

reduced by the existence of glacial ‘‘refuges’’ in Southern Europe during the Late Pleis-

tocene (Sommer and Nadachowski 2006).

Here we tested if climatic cycles and the long-term climatic changes superimposed on

them increased species accumulation rates over time. Still, we tested what an effect refuges

had on SAR slopes in the Late Pleistocene, and how the factors time and space intermingle

in determining paleospecies diversity across scales. Understanding how ecological models

of species accumulation over space and time describe the course of the species accumu-

lation curve in Plio-Holocene faunas would give new and fundamental insights as of

diversity change in the wake of the strongest climatic oscillations.

Materials and methods

The European Plio-Holocene large mammals record

Our analyses were performed by using an incidence (1/0) matrix of large mammal species

occurrences distributed across 685 fossil localities (local faunal assemblage, LFA) of

Western Eurasia spanning in age from the middle Pliocene (some 3.8 My) to the Early

Holocene (ca. 5,000 ky). For each locality we recorded, besides its species list, the latitude

and longitude, and calculated a numerical age estimate via temporal ordination techniques

(e.g. Alroy 2000; Fortelius et al. 2006, see Raia et al. 2009). Temporal ordination is a

statistical procedure that calculates a vector of numerical age estimates (one for each LFA)

based on species appearance events (Alroy’s ‘‘appearance event ordination’’ method) or

based on species-list similarity between LFAs (Fortelius et al.’s ‘‘spectral ordering’’).

We included only large bodied species (estimated body weight [ 5 kg) belonging to the

order Perissodactyla, Artiodactyla, Carnivora (with the member of family Ursidae, Cani-

dae, Hyenidae and Felidae), and Proboscidea. Smaller mammals were excluded because

their fossilization potential is much smaller (Damuth 1982; Rodrıguez et al. 2004; Raia

et al. 2006a). Our database includes 685 LFAs and 184 species. The full dataset is available

in Raia et al. (2009).

LFA geographic coordinates were imported in ArcGis 9.2 (ESRI) to produce a shapefile

recording their geographical distribution. As we dealt with a wide geographic area, we

used a Lambert Azimuthal Equal Area projection to avoid distortion of area due to the

variation of world’s curvature radius along meridians. We extracted a grid containing cells

of 500,000 Km2 (corresponding to squares of 707 km per side) from the LFA distribution

shapefile (Fig. 1), which minimizes the number of cells needed to include all LFAs. Then,

we calculated the actual landmass area within each cell of the grid (excluding the area

occupied by the sea, see Table 1) and counted the number of LFAs per cell.

174 Evol Ecol (2011) 25:171–188

123

Page 5

The landmass area per cell datum refers to its current value. Of course, this value

actually changes through the (geologic) time. Yet, sensible changes in Western Eurasia’s

geography are older than the Pliocene (e.g. Apennine orogenesis, Probosciedean Datum

Event, Messinian salinity crisis), hence outside the temporal interval considered here.

In addition, Plio-Pleistocene glaciations cyclically exposed some now submerged areas

by marine regression (e.g. the Northern stretches of the Adriatic Sea, several areas

in the North Sea), but none of the fossil sites included here lays underwater now.

Fig. 1 Geographic distribution of Plio-Holocene fossil sites (LFAs), yielding large mammal remains, usedin this study. The grid of geographic cells including LFAs is shown. For each cell, the identification numberis reported

Table 1 Cells number of LFA, species richness S, time span covered by the included LFAs, and actuallandmass area

CELL ID Duration (year) # LFAs S Area (Km2)

4 1,861,671 45 58 201,275

9 2,409,202 23 48 137,318

10 2,681,612 138 100 484,577

11 2,934,829 134 99 497,356

12 3,544,073 48 60 500,000

13 3,395,000 32 56 440,619

14 3,400,708 30 78 474,891

17 2,313,024 52 96 399,007

18 3,150,152 37 73 148,549

19 3,809,344 88 90 231,177

20 2,685,000 36 81 428,088

22 1,775,000 15 48 359,498

Evol Ecol (2011) 25:171–188 175

123

Page 6

Therefore, we consider that our estimation of cell surface area provides a good estimation of

the area occupied by the communities included in the analyses. Cells containing either less

than 7 LFAs, or whose LFAs cover a temporal interval less than 1.7 My long, were

excluded. We had empirically estimated that these criteria maximize the sample size

(=number of included cells) while maintaining high temporal coverage per cell. In total, 12

cells were retained and used as units of reference in order to produce SARs and STARs. As

our study is exploratory in essence, we avoided to define a priori specific biogeographical

provinces for which computing SAR (cfr. Barnosky et al. 2005). By using cells, we relied on

a restricted standardised criterion as for both the temporal and the spatial intervals to use.

STR, SAR and STAR calculation

We used three different statistical models of species accumulation (STR, SAR, and STAR)

representing two factors (time, space, and eventually their interaction) and all affected by

the same sampling noise (sampling inequality of LFAs over time and space). A first STR

model was fitted at the largest spatial scale (that is 12 selected geographical cells spanning

over the entire Western Eurasia subcontinent, see Fig. 1) by computing the increase in total

sampled species number (S) locality after locality, after LFAs had been collated in chro-

nologic order according to their numerical age estimates (see Appendix 1—Electronic

supplementary material). For each couple of consecutive LFAs (which are points in time)

we recorded cumulative S starting from the older LFA and the time span intervening

between the two. When two localities had the same age estimate, we took the cumulative

S value combining them as they were a single LFA. In this perspective, our species-time

curves resemble the computation of SAR across islands and our data points are ‘islands in

time’ (Hadly and Maurer 2001). We did not use the ‘‘moving window’’ (or complete-

nested, see Carey et al. 2007) approach that is familiar in most STR computations (White

2007) for it is inappropriate here (time-intervals between LFAs are irregular and some

LFAs have the same numerical age, therefore cannot be averaged). The STR is not devoid

of sampling problems, because the fossil record is obviously discontinuous, and younger

time intervals are represented by disproportionally more LFAs (Raia et al. 2006b, 2009).

To correct for this sampling inequality we computed a second STR as follows: we divided

our data set in 100 ky-long time bins. Thereby, 39 time bins were obtained, since the oldest

LFA is 3.73 My old. Then, we randomly selected one LFA per time bin and computed the

STR slope and log-likelihood. One-thousand random STRs were produced this way; the

average slope, and the maximum likelihood estimated slope were then computed and

compared to the w-exponent obtained by fitting a single STR to the entire database.

Besides problems in sampling inequality, the fossil record is plagued by taphonomic

biases. Preferential selection of some prey by predators, temporally and geographically

uneven effect of human hunting, differential conditions for preservation of, for instance,

large versus small specimens due to bone weathering and grain of the including sediments

are all relevant taphonomic factors. These factors may alter the distribution of species

abundances to some extent (but not dramatically, see Jernvall and Fortelius 2002; Raia

et al. 2006a) for instance by producing overrepresentation of some prey species or of larger

species (Damuth 1982). Yet, for STR or SAR computation either, it just matters whether a

given species is represented or not. The worst problem, then, could be the delay in the

(recorded) species first appearance event, which is true for any fossil record anyway, but

should not be particularly relevant with a fine-grained resolution as ours. By computing

STR and SAR this delay is not of an impact on the slope value (as it just produces a shift

along the 9 [time] axis), unless the delay in some (worse sampled) periods is larger than

176 Evol Ecol (2011) 25:171–188

123

Page 7

in others. Since we have oversampling of younger faunas, the delay in species appearance

events should be lower for the latter, thereby increasing the calculated slope values. Yet, as

we decided to select only cells with similar duration, even this (hypothetical) issue, should

not be very problematic here.

The SAR model was similarly estimated by using all LFAs: we combined the cells into

random groups of 1, 2, 4, 8, and 12 cells, respectively, by exploring all possible combi-

nations of cells per group. For instance, with 12 cells there are 12 possible combinations of

group size = 1 cell, 66 possible combinations of group size = 2 cells, 495 possible

combinations of group size = 4 cells and so on. For each group, we computed a mean Sand a mean area, averaging over all possible combinations of cells. Logged S and area

averages were then regressed to each other. Our SAR sampling scheme is not nested, and

corresponds to type IIB curves in Scheiner (2003). In spite of the discussion about the

appropriateness of Scheiner’s classification to represent the differences between species-

area and species-accumulation curves (see Gray et al. 2004; Whittaker and Fernandez-

Palacios 2007), here we prefer to use it for the clarity of its definition in regard to the

accumulation processes we study. By taking the averages of all possible combinations we

could, at least partially, account for the uneven distribution of LFAs over time period and

over space (Raia et al. 2009).

Besides using all possible combinations of cells, we also computed SAR slopes by

extracting all the LFAs belonging to a single cell which are included in a given 500 ky-

long time interval. Data from consecutive time bins were then aggregated as to compute

the SAR. This procedure assumes that all the information available for each time bin is

limited to a single, randomly chosen, cell. One thousand SARs were produced this way.

The average z-exponent, and the maximum likelihood estimated z-exponent were then

computed.

There is strong evidence that the climatic signature (if any) on community evolution

differs between southern and northern faunas, in Europe (Rodrıguez 2006). To model these

possible differences, we calculated the median of the LFAs latitudes and partitioned the

record in two halves, one including LFAs staying north to the median (the ‘‘North’’

sample) the other including LFAs staying south to it (the ‘‘South’’ sample). SAR and STR

calculations, performed at different spatial and temporal scales, were repeated on the

‘‘North’’ and ‘‘South’’ subsets. We expect STR slopes should be steeper in the South,

because southern faunas hosted both typically warm-adapted taxa and cold-adapted taxa

(when glaciers advanced southwards because of cold phases), whereas Northern faunas

should typically include cold-adapted species only (Lister 2004). Hence, turnover ought to

be higher (in time) in the South. As of SAR slopes, we expect they are steeper in the South

because of possible regionalization of faunas distributed in the major South-European

peninsulas, and of reported homogeneity of Pleistocene high-latitude environments

(Guthrie 2001).

Finally, we computed the STAR model as in Adler and Lauenroth (2003) to estimate its

ability to predict richness at various combinations of temporal and spatial extents, and to

know how time and space interact with each other. Samplings over time were repeated at

increasing temporal intervals, from the oldest LFA to: 3 million years ago (3.0 My), 2.0,

1.0, 0.5, 0.1 and finally to 0.01 My. This means the first temporal interval includes all

LFAs older than 3 My, than the second includes all LFAs older than 2 My, the third

includes all LFAs older than 1 My and so on. For each temporal interval, spatial samplings

were repeated at spatial scales of 1, 2, 4, 8, and 12 cells (=all selected cells), respectively

(see Table 2). Since spatial scales of \12 cells comprehend a number of possible com-

binations of cells, we calculated a single S value per spatial scale by averaging across all

Evol Ecol (2011) 25:171–188 177

123

Page 8

possible S (one for each combination of cells) as described above for the computation of

SAR. We assumed the minimum cell area and the minimum time span between successive

temporal sampling intervals as to represent the units of area and time, respectively. The

STAR model was then fitted to the data. A second STAR was calculated adding a fifth term

(the logarithm of the total number of LFAs at given space and time parameters) to the

equation, to correct for the unequal number of sites contributing to any given S value (for

instance, the number of LFAs per cell is highly correlated to its total species richness;

n = 12, r = 0.798; P = 0.002). This second STAR model takes the form:

log S ¼ log cþ z1 log Aþ w1 log T þ uðlog AÞðlog TÞ þ b log nLFAs

The coefficient u describes the degree of interaction (positive or negative, either)

between area and time.

A direct inspection of this interaction requires computing both SAR and STR exponents

at different temporal and spatial scales, respectively. If the interaction between area and

time on S is negative and linear, SAR exponent z must decline by sampling longer temporal

intervals, and STR exponent w must decline by sampling over larger spatial scale. For

SAR, we used the temporal intervals described above. As for STR, we calculated average

w obtained by testing all possible combinations of 1, 2, 4, 8, and 12 cells.

STR, SAR and STAR calculations at different scales were implemented in R.

Comparing STR slopes at different time periods

We tested the hypothesis that the STR slope changed in time, which is expected as a

consequence of the intense Quaternary cooling trend.

Five consecutive, non-nested STRs were produced using the temporal sampling points

described above as a reference. That is, one STR was computed over the 3.8–3 My

Table 2 (A) Average Species richness S and number of LFAs (B) at various temporal and geographicalscales, used to compute STAR models. At each temporal and spatial interval, the actual interval length(=actual mean time elapsed, in years) and size (=actual mean area, in km2) are reported

Mean area # ofcells

Mean timeelapsed (year)

736,334 1,801,901 2,806,832 3,238,737 3,717,446 3,821,736

Samplingpoint intime (Ma)

3.0 2.0 1.0 0.5 0.1 0.01

A, mean S

358,529 1 11.50 17.30 29.75 36.33 70.75 73.09

717058.8 2 15.40 20.44 44.95 55.85 86.53 101.03

1,434,118 4 21.00 23.81 66.07 82.93 119.06 130.42

2,868,235 8 – 41.58 92.41 116.42 153.77 162.22

4,302,353 12 – 52.00 112.00 140.00 177.00 184.00

B, mean # of LFAs

358,529 1 4.80 5.78 10.18 11.27 20.09 56.50

717058.8 2 5.20 6.67 18.18 22.55 36.83 113.00

1,434,118 4 10.40 12.00 36.36 45.69 73.67 226.00

2,868,235 8 – 24.00 72.73 90.18 147.33 452.00

4,302,353 12 – 30.00 100.00 124.00 221.00 678.00

–, not enough cells

178 Evol Ecol (2011) 25:171–188

123

Page 9

interval, the second over the 3–2 My interval and so on. For these analyses, we decided to

exclude LFAs younger than 100 ky, in order to avoid biases that might be introduced by

either human activity (Pushkina and Raia 2008) or by the very high number of LFAs

younger than 100 ky in our dataset (Raia et al. 2009).

The STR w exponents are expected to be steeper in younger intervals (when climate

variability increased, White et al. 2006). Two statistical comparisons of w exponents were

computed. First, w differences were tested by ANCOVA. Then, all five STRs were forced

through the origin. Regression through the origin is theoretically permitted here because

when the increment in time sampled is 0 the increment in the number of species sampled

must be zero. Yet, residuals calculated by regressing through the origin do not sum to zero,

and the R-squared value cannot be directly compared to a model with the intercept. Since

ordinary least squares (OLS) regression is not advisable when forcing the model through

the origin because the error term is transferred to the dependent variable, we used stan-

dardized major axis (SMA) regression when STR models were fitted through the origin

(Keough and Quinn 2002). A fundamental property of regression through the origin is that

it ignores the intercept. In both SAR and STR the intercepts must be viewed as the species

richness at the unitary scales, but it is well known they influence the curvature (slope) of

the relationships as well (Rosenzweig 1995). Ignoring ‘‘standing’’ species richness is

particularly desirable here since this figure is greatly influenced by sampling intensity.

Since the latter is much greater in younger time intervals, we might have introduced a

systematic bias along the time axis by using regression models with intercepts. Therefore,

we decided to use regression through the origin as a complementary test for the difference

in w exponents, although we did not append any special significance to the exponents

themselves or to the R-squared values.

Results

Species richness per cell S varies in between 36 and 102 species (mean = 73.1; see

Table 1).

STR takes the form, S(T) = 2.48E-06 T1.190 (F = 16,243.4, nLFA = 677, R2 = 0.960,

P \ 0.001, 95% CI around the slope: 1.109–1.304, see Fig. 2a). SAR takes the form

S(A) = 0.70A0.337 (F = 126.7, n = 5, R2 = 0.992, P \ 0.001, 95% CI around the slope:

0.313–0.424, see Fig. 2b).

In both STAR models the z1 exponents we calculated (0.169 and 0.192) are much lower

than z obtained by fitting SAR (Table 3). The STAR scaling exponents w1 (1.00 and 0.680)

are similar to those reported by Rosenzweig (1998) for STRs calculated over the

Fig. 2 Species-time relationship plot (a). Specie-area relationship plot (b). The x-axis represents the log ofnumber of years sampled (a) and the log of the surface area sampled in Km2 (b). The y axis represents thelog of the number of species

Evol Ecol (2011) 25:171–188 179

123

Page 10

evolutionary time, and both lower than w calculated by the STR model. The corrections for

the number of LFAs recorded at various combinations of time and space consistently

reduce the w1 but not the z1 slope, indicating that species accumulation over time is more

sensible to space than vice versa (Table 2). The interaction coefficient u is negative when

correcting for sampling richness yet it is positive when this correction is not applied. Both

STAR models perform very well in predicting species richness, R square is 0.921 for the

model with sampling correction and 0.954 for the uncorrected model (Fig. 3).

The maximum likelihood estimate of the SAR exponent obtained by extracting, for

consecutive 500 ky long time bins, all LFAs included in a single cell chosen at random is

0.274 (log-likelihood = 28.125). The average z-exponents over these randomizations is

0.281 (CI = 0.192–0.734, n = 1,000). The relationship between SAR exponent and

temporal extent of sampling is non-linear (Table 4). Exponents z increase to a sampling

interval of some 3 My, that is when the middle Pleistocene is included, but then decrease

when longer intervals (up to 3.8 My) are sampled (Table 5). Conversely, STR exponents

increase consistently with sampled area (Table 4). These results suggest the interaction

between area and time on species diversity is non-linear, and a large positive u value, as

obtained in the uncorrected STAR model is not reliable. Additionally, random STRs

obtained by sampling one LFA each 100 ky gave an average w of 0.810 (95% CI 0.740–

0.879) over 1,000 randomizations. The maximum likelihood estimate calculated over these

randomizations is 0.777 (log-likelihood is 29.688 for this particular STR); suggesting that

the very steep slope we obtained by fitting STR model to the entire data-set is most

probably influenced by overrepresentation of younger LFAs.

STR slopes per time

Either by using regression through the origin or fitting regression with an intercept term,

the change in STR slopes through time is far from linear when OLS regression is used

Table 3 Parameters of theSTAR models

Model parameters Without correction With correctionfor # of LFAs

c 1.01 0.807

z1 0.169 0.192

w1 1.004 0.680

u 0.445 -0.091

b 0.334

Fig. 3 Species-time area relationship (STAR) plots. a STAR with no correction for sampling inequalityacross scales. In b the STAR is computed correcting for the number of LFAs included at each samplingscale. Predicted and observed species richness (S) are reported in the x and y axis, respectively

180 Evol Ecol (2011) 25:171–188

123

Page 11

Ta

ble

4E

xp

on

ents

of

the

SA

Rm

od

elth

roug

hti

me

(z-

exp

onen

t)an

do

fth

eS

TR

mod

elth

rou

gh

spac

e(w

exp

on

ent)

To

tal

Tem

po

ral

sam

pli

ng

po

int

3.0

Ma

2.0

Ma

1.0

Ma

0.5

Ma

0.1

Ma

0.0

1M

a

z-E

xp

on

ent

0.4

75

(0.2

61–

0.7

46

)0

.344

(0.2

64–

0.4

32

)0

.620

(0.5

49–

0.7

05

)0

.683

(0.6

05

–0

.78

0)

0.4

71

(0.4

23

–0

.52

3)

0.3

67

(0.3

08

–0

.42

7)

Gro

up

size

(#o

fce

lls)

24

81

2

w-E

xp

on

ent

0.9

78

(0.1

60–

1.9

01

)1

.041

(0.6

30

–1

.45

4)

1.1

07

(0.8

03

–1

.34

5)

1.1

90

(1.1

09

–1

.30

4)

No

rth

z-E

xp

on

ent

0.5

64

a0

.274

a0

.618

(0.5

54–

0.6

83

)0

.675

(0.5

57

–0

.79

2)

0.4

11

(0.3

45

–0

.47

7)

0.3

63

(0.3

07

–0

.42

0)

w-E

xp

on

ent

0.8

33

(0.4

32–

1.9

91

)0

.924

(0.4

09

–2

.02

0)

0.8

84

(0.6

47

–1

.21

3)

0.8

18

(0.7

88

–0

.84

9)

So

uth

z-E

xp

on

ent

1.1

49

a0

.542

(0.4

20–

0.6

66

)0

.449

(0.3

50–

0.5

49

)0

.628

(0.5

06

–0

.74

9)

0.9

87

(0.8

03

–1

.16

8)

0.4

36

(0.3

49

–0

.52

4)

w-E

xp

on

ent

1.3

32

b

(0.1

55–

4.8

39

)1

.104

(0.1

30

–1

.86

5)

1.2

18

(1.1

62

–1

.26

8)

1.3

09

(1.2

62

–1

.35

6)

Co

nfi

den

cein

terv

als

are

rep

ort

edin

par

enth

eses

bel

ow

the

slop

es.

Fo

rb

oth

anal

yse

s,w

eca

lcu

late

dex

po

nen

tsb

yu

sin

gth

een

tire

reco

rd(t

ota

l);

and

then

repea

ted

the

anal

yse

so

nth

ere

cord

par

titi

on

edin

on

e‘‘

No

rth

’’an

do

ne

‘‘S

outh

’’h

alv

es,

repre

sen

tin

gL

FA

slo

cate

dei

ther

no

rthw

ard

or

sou

thw

ard

toth

eL

FA

sm

edia

nla

titu

de,

resp

ecti

vel

ya

Bas

edon

less

than

5ce

lls

bB

ased

on

two

com

bin

atio

ns

on

ly

Evol Ecol (2011) 25:171–188 181

123

Page 12

(Table 6; Fig. 4). Whereas both models support steep slopes since 0.5 My ago, results are

controversial for older time periods. OLS regression (with intercept) suggests that the rate

at which cumulative species increased was quite higher in the 3.0–2.0 My interval than

Table 5 Comparisons of SAR slopes calculated at different temporal scales. In the last column, H0 = SARexponents do not differ from zero

Time-interval(in My)

R2 Slope CI P-values, H0 = slopes do not differ per time period Pslope = 0

3.8–3.0 3.8–2.0 3.8–1.0 3.8–0.5 3.8–0.1 3.8–0

3.8–3.0 0.936 0.475 0.246–0.703 1 0.183 0.145 0.073 0.956 0.198 0.007

3.8–2.0 0.933 0.343 0.261–0.426 0.183 1 0.001 0.001 0.011 0.596 \0.001

3.8–1.0 0.973 0.619 0.542–0.697 0.145 0.001 1 0.252 0.006 0.001 \0.001

3.8–0.5 0.973 0.683 0.597–0.769 0.073 0.001 0.252 1 0.002 0.001 \0.001

3.8–0.1 0.978 0.470 0.421–0.52 0.956 0.011 0.006 0.002 1 0.006 \0.001

3.8–0 0.992 0.366 0.307–0.425 0.198 0.596 0.001 0.001 0.006 1 \0.001

Table 6 SMA through the origin and OLS with intercept regression models, used to fit STR equation for 5consecutive time intervals. H0 = slope does not differ between time periods

Time-interval(in My)

n R2 Slope CI P-values, H0 = slopes do not differ per time period

3.8–3.0 3.0–2.0 2.0–1.0 1.0–0.5 0.5–0.1

SMA through the origin

3.8–3.0 12 0.991 0.215 0.203–0.229 1 0.845 0.001 0.003 0.009

3.0–2.0 16 0.990 0.214 0.202–0.226 0.845 1 0.001 0.002 0.002

2.0–1.0 67 0.997 0.275 0.271–0.279 0.001 0.001 1 0.001 0.001

1.0–0.5 22 0.995 0.241 0.233–0.249 0.003 0.002 0.001 1 0.001

0.5–0.1 89 0.993 0.262 0.258–0.267 0.009 0.002 0.001 0.001 1

OLS regression with intercept

3.8–3.0 12 0.913 0.651 0.509–0.793 1 0.075 0.015 0.032 0.001

3.0–2.0 16 0.901 0.835 0.676–0.99 0.075 1 0.001 0.001 0.001

2.0–1.0 67 0.830 0.430 0.382–0.478 0.015 0.001 1 0.278 0.001

1.0–0.5 22 0.916 0.474 0.407–0.541 0.032 0.001 0.278 1 0.001

0.5–0.1 89 0.952 1.481 1.410–1.551 0.001 0.001 0.001 0.001 1

Fig. 4 Species-time relationships (STRs) slopes (y-axis) computed for 5 consecutive, not nested temporalintervals (reported in the x-axis). In a STRs are computed by standardized major axis regression through theorigin. In b STRs are computed by ordinary least squares (OLS) regression with intercept

182 Evol Ecol (2011) 25:171–188

123

Page 13

in the succeeding 1 million year. Conversely, SMA regression (through the origin) indi-

cates that STR slopes were generally higher since 2 My (Table 6; Fig. 4).

Discussion

Species accumulation in time

Our STR slopes (both w and w1) are significantly higher than ever recorded in living

species (0.2–0.4 in Connor and McCoy 1979; and see Rosenzweig 1995, 1998). Yet,

according to Preston’s conjecture, models built on very large time intervals should show

steeper curves (Preston 1960; Rosenzweig 1995, 1998; McKinney and Frederick 1999;

Adler and Lauenroth 2003). This happens to be the case because over the evolutionary time

(our time span is as long as 3.8 My) speciation adds to the cumulative species richness

curve (which on the short-time ecological scale just depends on immigration of species

from surrounding areas). STR slope changes a little with sampling area (mean w changes

from 0.98 to 1.2 over a 12-fold increase in area, Table 4, and always with large confidence

intervals). The evidence that STR slope was not constant over time is robust (Table 6). Yet,

the notion that w increases linearly with time is controversial, depending on the regression

model applied. For older time periods, it is possible that the temporal spacing between

LFAs (which is definitely larger than for younger periods, see Appendix 1—Electronic

supplementary material) makes temporal turnover to appear more abrupt, and w to be

steeper as an artefact. If this is true, better sampling of younger time periods would produce

shallower slopes in these latter intervals. However, the increase in w over the past 500 ky is

supported by both regression methods and is probably genuine. Global climatic variability

became very intense during this period. For instance, mean annual surface temperatures in

Antartica were changing as much as 15�C from glacial to interglacial periods (Jouzel et al.

2007). The longest (and one of the warmest) interglacial, corresponding to Marine Isotopic

Stage (MIS) 11 occurs in this period (Raynaud et al. 2005), as does the coldest glacial

phase (MIS 2). The STR is an indirect proxy for temporal turnover (Rosenzweig 1998;

White 2004). As such, our results suggest that the pace of community evolution accelerated

during late Pleistocene. Elsewhere we got clear evidence that climatic changes controlled

temporal turnover, and that turnover rate peaked in the late Pleistocene (Raia et al. 2005;

Meloro et al. 2008). However, those studies were drawn at a much smaller spatial scale

(peninsular Italy vs Western Eurasia) than the present work. It is possible that the climatic

signature on faunal evolution in Italy is that evident for the latter behaved as a glacial

‘‘refugium’’ (Sommer and Zachos 2009), hosting continuous change in community com-

position, but not in ecological structure (Rodrıguez 2006), as the ice sheets retreated and

advanced in keeping with the glacial/interglacial cycles. This would indicate that the

chance of finding a climatic signature on community evolution is scale-dependent

(Barnosky 2001), and varies from place to place (Rodrıguez 2006). In keeping with this

contention, we found much higher STR slopes in southern, than in northern, faunas

(see Table 4).

Species accumulation over space

The SAR slopes are much closer to the values found for both living species and pale-

odiversity data computed within biogeographical provinces (that is in the 0.1–0.3 range,

Connor and McCoy 1979; Barnosky et al. 2005). Indeed, fitting a SAR model we got

Evol Ecol (2011) 25:171–188 183

123

Page 14

an exponent of 0.34, yet this value is probably influenced by the large temporal scale

sampled, for both z1 exponents obtained by STAR models are much shallower. The

maximum likelihood estimate of z calculated by sampling one geographic cell per 500 ky

long time bin is 0.28, which is higher than z1s as well. Barnosky et al. (2005) obtained

shallower SAR slopes when paleodiversities were corrected for sampling bias by means of

rarefaction analysis. Yet, they did not account for the effect of time. Arguably, speciation,

immigration and climatic changes (and their interaction) have much less of an influence on

z than on w here for a number of reasons. Firstly, we analyzed mammal species whose

geographic ranges are much larger than Europe. Mammoths, horses, bison, aurochs, red

and fallow deer, roe deer, cave lion, leopard, wolf, cave hyena and cave bears occurred, to

name just a few renewed examples, over most of Asia, Africa, and even North America (in

a minority of cases). If this indicates climatic tolerance in these species, spatial turnover

could not have been very high anyhow. By restricting ourselves to large species we got

much larger ranges than for all mammals, hence much more similarity between local

faunas and shallower slopes than we would get by sampling smaller species as well.

Second, habitat heterogeneity is known to affect SAR slope (more diverse habitats have

higher spatial turnover, thereby produce higher SAR slopes, Kallimanis et al. 2008), but

there is evidence that habitat heterogeneity (in the geographical area we considered)

decreased during late Pleistocene. For instance, late Pleistocene environments in Northern

Europe were dominated by the so-called mammoth steppe, a homogeneous steppic biome

that extended from Western Eurasia to Kamchatka (Guthrie 2001). Not surprisingly, SAR

slopes calculated for Northern faunas appear much shallower during late Pleistocene times,

when the mammoth steppe was established (Table 4). In a similar vein, the high slopes in

the North we recorded for the 1–0.5 My time interval (Table 4) make sense when we

consider that during this time period it took place a massive immigration of new species

from Asia through Northern and Eastern Europe, thereby establishing what it is known as

the ‘‘Galerian’’ mammal age (Raia et al. 2009) in Europe. This immigration event is

recorded by SAR as well. When SAR is computed including the beginning of this period,

the z exponent becomes much steeper (Table 5). Thirdly, during the Ice Ages many species

are known to have moved repeatedly along a North-South vector in response to climatic

shifts (migrating southward when glaciers advanced in cold periods and vice versa during

more favorable climatic regimes, Lyons 2003; Lister 2004; Sommers and Zachos 2009). As

our SAR was computed over the whole continent, and over a temporal span including

several migratory events, shallow z are expected. Suppose, for example, to compute a SAR

counting species over an interval including both a glacial period and then an interglacial

phase. Now, consider a group of steppe (Northern) adapted species. These species will not

be counted in the South over a short phase in the interglacial, and not in the North during

the glacial. However, they will be counted in both the North and in the South over a longer

interval. If it was possible to compute SAR at a much shorter temporal intervals, we would

have got much steeper z. It is interesting to note, though, that SAR slopes computed for

older time periods are generally higher than for the late Pleistocene (Table 4). This sug-

gests higher regionalization of faunas during early and middle Pleistocene, at least at the

temporal resolution we used here. Although we could not dismiss this result to depend on

sampling inequality, it is in agreement with a previous study of ours (Raia et al. 2009)

indicating that there were at least two different and geographically disjunctive paleo-

communities in Western Eurasia throughout early to middle Pleistocene, but only one

distinct assemblage of species in younger (late Pleistocene) times.

184 Evol Ecol (2011) 25:171–188

123

Page 15

Combining the effects of space and time

As for the estimation of diversity across scales, STAR models provide us with very good

estimates of cumulative species richness. The interaction factor u is negative in the STAR

model controlling for LFA richness (a measure of sampling intensity), which we argue

should be preferred over the uncorrected model. A negative interaction has been reported

for living species as well (Adler and Lauenroth 2003; Adler et al. 2005). This fact confirms

Preston’s second hypothesis (Preston 1960) that the rate of species accumulation over time

must decline as the sampled area increases and vice versa. The classic explanation for this

negative interaction is that at larger spatial scales a higher proportion of species in the

regional pool is identified over short time intervals, thereby reducing temporal species

accumulation rate (Adler and Lauenroth 2003; White 2004; White et al. 2006). This may

well be the case with Ice Age Western Eurasian large mammal faunas, which occurred in

two distinct paleocommunities during the Pliocene to middle Pleistocene period, and were

diachronous in the late Pleistocene because of massive and rapid intracontinental migration

along the North-South direction.

According to Vrba’s ‘‘turnover pulse’’ hypothesis, temporal taxonomic turnover is

buffered by (modest) habitat tracking and favours habitat specialization under ordinary

climatic oscillation, whereas more intense climate change disrupts community stability and

drives extinction and speciation events (Vrba 1995a, b). The first scenario should increase

SAR slopes, provided the calculation of SAR is framed at the proper scale. The second

scenario contributes to the increase in STR slopes. In keeping with these theoretical

expectations, we found z exponent was steeper during Pliocene to middle Pleistocene

times, when climate change was not as nearly intense as during the late Pleistocene, when

we record steeper w exponent.

Conclusion

Models built to describe species accumulation over space and over time at ecological

scales proved useful over the geologic time scale as well. Here, for Plio-Holocene large

mammals of Western Eurasia, we found that species accumulation rate over time was

much higher than over space, and was at least in part influenced by climatic changes. The

high w exponents we got are deemed to account for evolutionary phenomena not included

in STRs computed over the shorter –ecologic- temporal scale. This ‘‘third phase’’ of

species-time relationships show very high scaling exponents, as once found by Rosenzweig

(1998), although it is clear that the exponent itself is sensitive to the particular time period

considered for STR computation. There is evidence that space and time interact negatively

in determining species accumulation rates, in keeping with the findings of Adler and

Lauenroth (2003) and Adler et al. (2005). Consequently, particular attention to scale must

be given in any account of paleodiversity as already pointed out by Barnosky (2001) and

Barnosky et al. (2005).

Intra-continental dispersal is known to have been massive during the late Pleistocene.

Dispersal has probably increased community survival providing a limit to the influence of

climatic change effects at the continental scale. The greatly reduced opportunity for dis-

persal in modern faunas may prove very problematic for living species, in the light of

current climatic change.

Evol Ecol (2011) 25:171–188 185

123

Page 16

Acknowledgments Shai Meiri and Anna Loy provided us with important comments and advice that let usincreasing the quality of this manuscript. Anastassios Kotsakis read an earlier version of the manuscript andhelped us collecting and preparing the database used for this study. We are grateful to two anonymousreviewers for their constructive comments on this manuscript.

References

Adler PB, Lauenroth WK (2003) The power of time: spatiotemporal scaling of species diversity. EcologyLett 6:749–756

Adler PB, White EP, Lauenroth WK, Kaufman DM, Rassweiler A, Rusak JA (2005) Evidence for a generalspecies-time-area relationship. Ecology 86:2032–2039

Alroy J (2000) New methods for quantifying macroevolutionary patterns and processes. Paleobiology26:707–733

Barnosky AD (2001) Distinguishing the effects of the Red Queen and Court Jester on Meiocene mammalevolution in the northern Rocky Mountains. J Vert Paleont 21:172–185

Barnosky AD, Kraatz BP (2007) The role of climatic change in the evolution of mammals. Bioscience57:523–532

Barnosky AD, Carrasco MA, Davis EB (2005) The impact of the species-area relationship on estimates ofpaleodiversity. Plos Biol 3:1356–1361

Bobe R, Behrensmeyer AK (2004) The expansion of grassland ecosystems in Africa in relation tomammalian evolution and the origin of the genus Homo. Palaeogeog Palaeoclimatol Palaeoecol207:399–420

Carey S, Ostling A, Harte J, del Moral R (2007) Impact of curve construction and community dynamics onthe species-time relationship. Ecology 88:2145–2153

Connor EF, McCoy ED (1979) The statistics and biology of the species-area relationship. Am Nat 113:791–833

Damuth J (1982) Analysis of the preservation of community structure in assemblages of fossil mammals.Paleobiology 8:434–446

Dengler J (2009) Which function describes the species-area relationship best? A review and empiricalevaluation. J Biogeog 36:728–744

Fattorini S (2007) To fit or not to fit? A poorly fitting procedure produces inconsistent results when thespecies—area relationship is used to locate hotspots. Biodiv Conserv 16:2531–2538

Fortelius M, Gionis A, Jernvall J, Mannila H (2006) Spectral ordering and biochronology of European fossilmammals. Paleobiology 32:206–214

Gray JS, Ugland KI, Lambshead J (2004) On species accumulation and species-area curves. Glob EcolBiogeogr 13:567–568

Guthrie RD (2001) Origin and causes of the mammoth steppe: a story of cloud cover, woolly mammal toothpits, buckles, and inside-out Beringia. Quat Sci Rev 20:549–574

Hadly EA, Maurer BA (2001) Spatial and temporal patterns of species diversity in montane mammalcommunities of western North America. Evol Ecol Res 3:477–486

Harte J, Kinzig A, Green J (1999) Self-similarity in the distribution and abundance of species. Science284:334–336

Hubbell SP (2001) The unified neutral theory of biodiversity and biogeography. Princeton University Press,Princeton

Jouzel J, Masson-Delmotte V, Cattani O, Dreyfus G, Falourd S, Hoffmann G, Minster B, Nouet J, BarnolaJM, Chappellaz J, Fischer H, Gallet JC, Johnsen S, Leuenberger M, Loulergue L, Luethi D, Oerter H,Parrenin F, Raisbeck G, Raynaud D, Schilt A, Schwander J, Selmo E, Souchez R, Spahni R, Stauffer B,Steffensen JP, Stenni B, Stocker TF, Tison JL, Werner M, Wolff EW (2007) Orbital and millennialAntarctic climate variability over the past 800,000 years. Science 317:793–796

Kallimanis AS, Mazaris AD, Tzanopoulos J, Halley JM, Pantis JD, Sgardelis SP (2008) How does habitatdiversity affect the species-area relationship? Global Ecol Biogeog 17:532–538

Keough MJ, Quinn GP (2002) Experimental design and data analysis for biologists 520 pages. CambridgeUniversity Press, Cambridge

Kinzig AP, Harte J (2000) Implications of endemics-area relationships for estimates of species extinctions.Ecology 81:3305–3311

Lister AM (2004) The impact of Quaternary Ice Ages on mammalian evolution. Phil Trans Roy Soc B359:221–241

Lyons SK (2003) A quantitative assessment of the range shifts of Pleistocene mammals. J Mammal 84:385–402

186 Evol Ecol (2011) 25:171–188

123

Page 17

Lyons SK (2005) A quantitative model for assessing community dynamics of Pleistocene mammals. Am Nat165:E168–E185

Magurran AE (2007) Species abundance distributions over time. Ecology Lett 10:347–354May RM (1975) Patterns of species abundance and diversity. In: Diamond ML, Cody JM (eds) Ecology and

evolution of communities. Belknap Press, Cambridge, pp 81–120McKinney ML, Frederick DL (1999) Species-time curves and population extremes: Ecological patterns in

the fossil record. Evol Ecol Res 1:641–650Meloro C, Raia P, Carotenuto F, Barbera C (2008) Diversity and turnover of Plio-Pleistocene large mammal

fauna from the Italian Peninsula. Palaeogeog Palaeoclimat Palaeoecol 268:58–64Myers N, Mittermeier RA, Mittermeier CG, da Fonseca GAB, Kent J (2000) Biodiversity hotspots for

conservation priorities. Nature 403:853–858Pimm SL, Askins RA (1995) Forest losses predict bird extinctions in Eastern North-America. Proc Nat Acad

Sci USA 92:9343–9347Plotkin JB, Potts MD, Leslie N, Manokaran N, LaFrankie J, Ashton PS (2000) Species-area curves, spatial

aggregation, and habitat specialization in tropical forests. J Theor Biol 207:81–99Preston FW (1960) Time and space and variation of species. Ecology 41:611–627Pushkina D, Raia P (2008) Human influence on distribution and extinctions of the late Pleistocene Eurasian

megafauna. J Hum Evol 54:769–782Raia P, Piras P, Kotsakis T (2005) Turnover pulse or Red Queen? Evidence from the large mammal

communities during the Plio-Pleistocene of Italy. Palaeogeog Palaeoclimat Palaeoecol 221:293–312Raia P, Meloro C, Loy A, Barbera C (2006a) Species occupancy and its course in the past: macroecological

patterns in extinct communities. Evol Ecol Res 8:181–194Raia P, Piras P, Kotsakis T (2006b) Detection of Plio-Quaternary large mammal communities of Italy. An

integration of fossil faunas biochronology and similarity. Quat Sci Rev 25:846–854Raia P, Carotenuto F, Meloro C, Piras P, Barbera C, Kotsakis T (2009) More than three million years of

community evolution. The temporal and geographical resolution of the Plio-Pleistocene WesternEurasia mammal faunas. Palaeogeogr Palaeoclimat Palaeoecol 276:15–23

Raynaud D, Barnola JM, Souchez R, Lorrain R, Petit JR, Duval P, Lipenkov VY (2005) Palaeoclimatology:the record for marine isotopic stage 11. Nature 436:39–40

Rodrıguez J (2006) Structural continuity and multiple alternative stable states in Middle PleistoceneEuropean mammalian communities. Palaeogeog, Palaeoclim, Palaeoecol 239:355–373

Rodrıguez J, Alberdi MT, Azanza B, Prado JL (2004) Body size structure in north-western MediterraneanPlio-Pleistocene mammalian faunas. Global Ecol Biogeog 13:163–176

Rosenzweig ML (1995) Species diversity in space and time. Cambridge University Press, CambridgeRosenzweig ML (1998) Preston’s ergodic conjecture: the accumulation of species in space and time. In:

McKinney ML, Drake JA (eds) Biodiversity dynamics: turnover of populations, taxa, and communi-ties. Columbia University Press, New York, pp 311–348

Rosenzweig ML, Sandlin EA (1997) Species diversity and latitudes: listening to area’s signal. Oikos80:172–176

Scheiner SM (2003) Six types of species-area curves. Glob Ecol Biogeogr 12:441–447Shmida A, Wilson MV (1985) Biological determinants of species diversity. J Biogeog 12:1–20Sommer RS, Nadachowski A (2006) Glacial refugia of mammals in Europe: evidence from fossil records.

Mammal Rev 36:251–265Sommer RS, Zachos FE (2009) Fossil evidence and phylogeography of temperate species: ‘glacial refugia’

and post-glacial recolonization. J Biogeogr 36:2013–2020Triantis KA, Nogues-Bravo D, Hortal J, Borges PAV, Adsersen H, Fernandez-Palacios JM, Araujo MB,

Whittaker RJ (2008) Measurements of area and the (island) species-area relationship: new directionsfor an old pattern. Oikos 117:1555–1559

Turner WR, Tjørve E (2005) Scale-dependence in species—area relationships. Ecography 28:721–730Vrba ES (1995a) The fossil record of African antelopes (Mammalia, Bovidae) in relation to human evo-

lution and paleoclimate. In: Vrba ES, Denton GH, Partridge TC, Burckle LH (eds) Paleoclimate andevolution with emphasis on human origins. Yale University Press, New Haven, pp 385–424

Vrba ES (1995b) On the connections between paleoclimate and evolution. In: Vrba ES, Denton GH,Partridge TC, Burckle LH (eds) Paleoclimate and evolution with emphasis on human origins. YaleUniversity Press, New Haven, pp 24–48

White EP (2004) Two-phase species-time relationships in North American land birds. Ecology Lett 7:329–336

White EP (2007) Spatiotemporal scaling of species richness: patterns, processes, and implications. In: StorchD, Marquet PA, Brown JH (eds) Scaling biodiversity. Cambridge University Press, Cambridge,pp 325–346

Evol Ecol (2011) 25:171–188 187

123

Page 18

White EP, Adler PB, Lauenroth WK, Gill RA, Greenberg D, Kaufman DM, Rassweiler A, Rusak JA, SmithMD, Steinbeck JR, Waide RB, Yao J (2006) A comparison of the species-time relationship acrossecosystems and taxonomic groups. Oikos 112:185–195

Whittaker RJ, Fernandez-Palacios JM (2007) Island biogeography. Ecology, evolution, and conservation,2nd edn. Oxford University Press, Oxford

Zachos J, Pagani M, Sloan L, Thomas E, Billups K (2001) Trends, rhythms, and aberrations in globalclimate 65 Ma to present. Science 292:686–693

188 Evol Ecol (2011) 25:171–188

123