Download Temporal stability of European rocky shore assemblages

Oikos 000: 001–009, 2012
doi: 10.1111/j.1600-0706.2011.19967.x
© 2011 The Authors. Oikos © 2012 Nordic Society Oikos
Subject Editor: Matthew Bracken. Accepted 21 November 2011
Temporal stability of European rocky shore assemblages: variation
across a latitudinal gradient and the role of habitat-formers
F. Bulleri, L. Benedetti-Cecchi, M. Cusson, E. Maggi, F. Arenas, R. Aspden, I. Bertocci, T. P. Crowe,
D. Davoult, B. K. Eriksson, S. Fraschetti, C. Golléty, J. N. Griffin, S. R. Jenkins, J. Kotta,
P. Kraufvelin, M. Molis, I. Sousa Pinto, A. Terlizzi, N. Valdivia and D. M. Paterson
F. Bulleri ([email protected]), Dipto di Scienze Botaniche. Ecologiche e Geologiche, Univ. di Sassari, Via Piandanna 4, IT-0071
Sassari, Italy. – FB, L. Benedetti-Cecchi, M. Cusson, E. Maggi, I. Bertocci, Dipto di Biologia, Univ. di Pisa, CoNISMa, Via Derna 1,
IT-56126, Pisa, Italy. MC also at: Dépt des sciences fondamentales, Univ. du Québec à Chicoutimi, 555 Boulevard de l’Université, Chicoutimi,
QC, G7H 2B1 Canada. – T. P. Crowe, School of Biology and Environmental Science, Univ. College Dublin, Belfield, Dublin 4, Ireland.
– IB and F. Arenas, Laboratory of Coastal Biodiversity, Centre of Marine and Environmental Research (CIIMAR), Univ. of Porto, Rua dos
Bragas 289, PT-4050-123 Porto, Portugal. – R. Aspden, C. Golléty and D. M. Paterson, Scottish Oceans Inst., Univ. St Andrews, St Andrews,
Fife, KY16 8LB, UK. – CG and D. Davoult, UPMC Univ Paris 06, UMR 7144, Station Biologique de Roscoff, FR-29682 Roscoff, France.
– B. K. Eriksson, Dept of Marine Benthic Ecology and Evolution, Centre for Ecological and Evolutionary Studies, Univ. of Groningen,
Nijenborgh 7, NL-9747 AG Groningen, the Netherlands. – S. Fraschetti and A. Terlizzi, Dipto di Scienze e Tecnologie Biologiche ed
Ambientali, Univ. del Salento, CoNISMa, IT-73100 Lecce, Italy. – J. N. Griffin and S. R. Jenkins, Marine Biological Ass., Citadel Hill,
Plymouth, PL1 2PB, UK. SRJ also at: School of Ocean Sciences, Bangor Univ., Menai Bridge, Anglesey, LL595AB, UK. – J. Kotta, Estonian
Marine Inst., Univ. of Tartu, Mäealuse 14, EE-12618 Tallinn, Estonia. – P. Kraufvelin, ARONIA, Coastal Zone Research Team, Åbo Akademi
University and Novia University of Applied Sciences, Raseborgsvägen 9, FI-10600 Ekenäs, Finland. – M. Molis and N. Valdivia, Biologische
Anstalt Helgoland, Sect. Functional Ecology, Alfred-Wegener-Inst. for Polar and Marine Research, Marine station, Kurpromenade 201,
DE-27498 Helgoland, Germany. NV also at: Inst. de Ciencias Marinas y Limnológicas, Facultad de Ciencias, Univ. Austral de Chile,
Campus Isla Teja, Valdivia, Chile. – I. Sousa Pinto, Centre of Marine and Environmental Research (CIIMAR), and Dept of Biology,
Faculty of Sciences, Univ. of Porto, Rua dos Bragas 289, PT-4050-123 Porto, Portugal.
Compensatory dynamics, overyielding and statistical averaging are mechanisms promoting the temporal stability of natural
communities. Using the model of European intertidal rocky shore assemblages and collating 17 datasets, we investigated
how the strength of these stability-enhancing mechanisms varies with latitude and how it can be altered by the loss of
habitat-formers (e.g. canopy-forming macroalgae). Community stability decreased with increasing latitude, mostly as a
consequence of a greater synchronization of species fluctuations. Statistical averaging and overyielding (i.e. richness effect)
promoted stability, but their strength did not vary with latitude. An experimental removal of macroalgal canopies caused a
strengthening of the statistical averaging effect that was consistent across the latitudinal gradient investigated. Nonetheless,
the loss of canopies depressed stability by enhancing the synchronization of species fluctuations on southernmost shores,
while it had weak effects on shores at higher latitudes. Variation in life-history traits among canopy-forming species and/
or in prevailing environmental conditions across a gradient of latitude could underlie variable effects of habitat-formers on
species fluctuations. Our study shows 1) that the stability of intertidal assemblages and strength of compensatory dynamics
varies with latitude, 2) that canopy-forming macroalgae, exerting a strong control on understorey species, can influence the
strength of compensatory dynamics and 3) that biological forcing (i.e. facilitation) can be as important as environmental
forcing in enhancing the synchronization of species fluctuations.
Concerns over the consequences of biodiversity loss for
the ability of ecosystems to supply goods and services to
mankind have fuelled research on the diversity-stability
hypothesis (Tilman 1996, Chapin et al. 1998, McCann
2000). This has ultimately led to the recognition that community stability can be promoted by different mechanisms
(Lehman and Tilman 2000, Cottingham et al. 2001, Loreau
and de Mazancourt 2008, Isbell et al. 2009).
Asynchronous species dynamics, fostered by interspecific
competition and/or differential response to disturbances
or environmental stress, can sustain stability by generating
negative species covariances (Tilman 1999, Yachi and
Loreau 1999, Ives et al. 1999, McCann 2000). In addition,
when most species perform better in multi- than monospecific assemblages, temporal stability is enhanced via a
greater increase in the average of the aggregate property in
respect to its variance. This mechanism, generally referred to
as overyielding, occurs whenever the average of a community
property increases with species diversity (Lehman and Tilman
2000). Finally, the statistical averaging of fluctuations in the
1
abundances of individual species within a community can
lead to greater temporal stability in communities at increasing levels of species richness (statistical averaging or portfolio
effect; Doak et al. 1998).
The role of these mechanisms and, in particular, of species
fluctuations, on stability has been investigated in a variety of
terrestrial and aquatic systems, including lacustrine plankton, vascular plant communities, invertebrate and vertebrate
assemblages (Klug et al. 2000, Fischer et al. 2001, Romanuk
and Kolasa 2002, Romanuk et al. 2006, Houlahan et al.
2007, Vasseur and Gaedke 2007, Valone and Barber 2008).
These studies have revealed that a variety of factors, such as
species traits and trophic-level, environmental conditions,
resource availability and spatial and temporal scale investigated, can generate large context-dependency in the strength
of the mechanisms underlying community stability.
Little effort has, however, been made to assess how the
importance of mechanisms generating stability can vary
in space, within a given ecosystem. For instance, there is a
trend for increased strength in biological interactions, such
as interspecific competition (Connolly et al. 2001, Bertness
and Ewanchuk 2002), herbivory (Pennings and Silliman
2005, Coleman et al. 2006) or predation (Vermeij 1978,
Sanford et al. 2003), from high to low latitudes. The significance of asynchronous species fluctuations can, therefore, be
expected to vary with latitude, as a consequence of variation
in the strength of biological interactions and environmental
forcing (Campbell et al. 2011). Predicting the direction of
this change is, however, difficult, since the relative importance of weak versus strong interactions (Klug et al. 2000,
McCann 2000, Griffin et al. 2009), as well as that of
differential response among species to environmental fluctuations (Ives et al. 1999, 2000, Yachi and Loreau 1999,
Klug et al. 2000), in generating compensatory dynamics is
yet to be fully elucidated.
Another aspect that has been poorly explored is the role
that habitat-formers or ecosystem engineers (species that
create or modify habitats) can play in determining stability
levels. The presence of habitat-formers is a major driver of
the structure and functioning of species assemblages (Bruno
and Bertness 2001). By setting prevailing environmental
conditions and, in particular, dampening their oscillations,
habitat-formers can modulate the direction and strength
of inter-specific interactions (Bruno and Bertness 2001,
Kraufvelin et al. 2006, Callaway 2007, Brooker et al. 2008).
Thus, they can be hypothesized to have a large influence
on the mechanisms that promote stability. For instance,
by modifying species interactions and/or their response to
disturbances or environmental stress, habitat-formers could
directly govern the synchronization of species fluctuations.
In addition, through their pervasive effects on species identity and relative abundances, they might alter the contribution of statistical averaging and overyielding to stability by
modulating evenness (Doak et al. 1998, Cottingham et al.
2001, Li and Stevens 2010; but see Isbell et al. 2009). Alteration to species evenness in the presence of habitat-formers,
such as canopy-forming macroalgae, seems likely when they
are perennial and their abundance/biomass is disproportionately greater than that of associated species. Some studies
have reported greater stability for dominant than subordinate species (Lepš 2004, Steiner et al. 2005, Polley et al.
2
2007). Thus, as argued by Grman et al. (2010), a large abundance of particularly stable species might enhance community stability by a mechanism similar to the sampling effect
(Huston 1997).
We investigated patterns of variation in the stability of
intertidal rocky shore assemblages and underlying mechanisms along a latitudinal gradient in Europe. Furthermore,
we assessed how the mechanisms promoting stability could
be altered by the loss of canopy-forming macroalgae. We
focused on these habitat-formers since their role in maintaining intertidal biodiversity has been well-documented,
and they are reported to be declining on many temperate
rocky shores worldwide (North Atlantic: Jenkins et al. 1999;
Mediterranean: Benedetti-Cecchi et al. 2001, Airoldi and
Beck 2007; New Zealand: Lilley and Schiel 2006; California:
Thom and Widdowson 1978; Baltic Sea: Vogt and Schramm
1991). Specifically, we tested the hypothesis that the stability in the cover of sessile rocky intertidal species would vary
along a gradient of latitude as a consequence of variation
in the synchrony of species fluctuations, statistical averaging
and overyielding (i.e. richness effect sensu Cardinale et al.
2006). In addition, we tested the hypothesis that the effect
of macroalgal canopy removal on stability and underlying
mechanisms would vary between southern and northern
European rocky shores. In line with the above mentioned
trend for the strength of biological interactions to decrease
towards higher latitudes (Vermeij 1978, Connolly et al.
2001, Bertness and Ewanchuk 2002, Sanford et al. 2003,
Pennings and Silliman 2005, Coleman et al. 2006), Campbell
et al. (2011) have suggested that a weakening of the relationship between diversity and stability towards higher latitudes
might be generated by the prevalence of physical forcing
over biological forcing. In accordance, we predicted that the
removal of canopies would have little effect on the stability
of rocky intertidal communities at higher latitudes, where
physical forcing prevails. In contrast, at lower latitudes,
where biological interactions (either negative or positive)
play an important role in regulating the relationship between
diversity and stability, the removal of canopies should generate a large effect on stability.
Material and methods
Estimates of temporal variance in total cover and in cover
of individual taxa are required for analysing temporal stability of communities and underlying mechanisms. A total of
17 datasets were used to investigate latitudinal patterns in
the mechanisms promoting stability (Table 1). Species fluctuations can be synchronous at some timescales, but not at
others, so that long time-series enabling partitioning at varying scales are of paramount importance to identify discontinuities (Vasseur and Gaedke 2007, Downing et al. 2008,
Keitt 2008). The overall duration (about two years) and
sampling intervals (i.e. 2–6 months) of the studies included
in our analyses enable encompassing major environmental
fluctuations (i.e. seasonal). Thus, although we cannot provide any insight into longer term trajectories, the data used
in this study are adequate to describe short- to mid-term
oscillations in rocky benthic assemblages characterized by
relatively fast species turn-over.
Experimental datasets used to assess the effects of canopy removal on stability and underlying mechanisms are reported in brackets under the Dataset column; this numbering order is followed in
Fig. 2; the total number of sampling events is reported in brackets in the Sampling interval column. athese datasets had a spatial structure (replicated sites) and generated 3 subsets of data. Dominant
taxa are reported for the cases in which canopy-formers were absent (dataset 3, 4) or not dominant (dataset 6, 8). Canopy cover (%) refers to the time at which the experiment was shortly started. The
ratio between the cover of the canopy versus that of the understorey sessile assemblage ( 1 SE) is reported under the column Canopy/understorey. ‡Bertocci et al. 2010, Mar. Ecol. Prog. Ser. 414:
107–116; †Bertocci et al. 2010, Ecology 86: 2061–2067; ††Vaselli et al. 2008, Mar. Ecol. Prog. Ser. 364: 57–66. EC  Encrusting corallines; AT  Algal turfs.
2.14  0.28
1.27  0.44
2
2
7.6  1.39
2
6.00  1.17
2
0.34  0.22
1.61  0.30
0.78  0.02
0.79  0.07
0.68  0.04
∼ 93
∼ 43
–
–
∼ 98
∼ 12
∼ 88
∼4
∼ 69
∼ 80
∼ 84
∼ 95
∼ 89
Cystoseira amentacea var. stricta
Chondrus crispus, Mastocarpus stellatus
None/EC–AT
None/EC–AT
Cystoseira compressa
Sargassum muticum/EC–AT
Ascophyllum nodosum
Sargassum muticum/EC–AT
Fucus serratus/Fucus vesiculosus
Fucus serratus/Fucus vesiculosus
Fucus serratus
Fucus serratus
Fucus serratus
Italy
Portugal‡
Italy†
Italy††
Italy
Spain
France
UK
UK
Ireland
Germany
Germany
UK
1(1)
2(2)
3
4
5(3)
6
7(4)
8
9(5)
10(6)
11(7)
12(8)
13(9)
Ionian Sea (Med.)
Atlantic
Ligurian Sea (Med.)
Ligurian Sea (Med.)
Ligurian Sea (Med.)
Atlantica
North Atlantic
North Atlantica
North Atlantic
North Atlantic
North Sea
North Sea
North Atlantic
2006–2007
2006–2007
2001–2003
2003–2005
2006–2007
2002–2004
2006–2008
2002– 2004
2006 –2007
2006– 2007
2006 – 2007
2004 – 2007
2006 –2008
3–6 month (5)
2–5 month (5)
2–6 month (7)
2–6 month (7)
3–6 month (6)
4–6 month (6)
4–7 month (4)
6–8 month (5)
3–6 month (6)
3–6 month (7)
3–6 month (7)
3–6 month (7)
3–6 month (6)
m2
0.09
0.09 m2
0.01 m2
0.06 m2
0.09 m2
0.67 m2
0.25 m2
0.67 m2
0.09 cm2
0.09 m2
0.09 m2
0.09 m2
0.09 m2
Canopy/understorey
Canopy cover (%)
Plot size
Canopy-forming species/dominant taxa
Sampling interval
Sampling period
Region
Country
Dataset
Table 1. Details of datasets used in the correlative and experimental analyses, ordered from low to high latitudes.
A subset of nine datasets including the experimental
removal of canopy-forming macroalgae were further used
to test our hypothesis on the effects of the loss of this morphological group on stability and underlying mechanisms
(Table 1). Canopy-forming species varied among sites
(Table 1). Except for Site 4, where canopies were made of
Ascophyllum nodosum, Fucus serratus and F. vesiculosus were
the main canopy-formers on northern shores. In contrast,
species forming canopies were more variable at lower latitudes: Chondrus crispus and Mastocarpus stellatus on the coast
of Portugal, Cystoseira compressa in the Ligurian Sea and
Cystoseira amentacea var. stricta in the Ionian Sea.
In all cases, canopies were removed at the beginning
of the experiment, recruits were selectively removed thereafter with no further disturbance to other species and
sampling started about three months after the initial treatment. Canopy-forming species were removed by hand to
minimize disturbance to understorey assemblages. At the
beginning of the experiment, the cover of the canopy varied among sites (Table 1), but there was not a latitudinal
pattern (Spearman rs  0.05, p  0.90). In contrast, the
ratio between the cover of the canopy and that of understorey assemblages showed a marked trend to decrease
when moving from southern to northern shores (Table 1),
although this correlation was not significant (Spearman
rs  20.62, p  0.07).
The percentage cover of all macroscopic sessile taxa
( 1 mm) was quantified visually in each study by means
of grids divided into sub-quadrats, except for datasets 6 and
8 where it was estimated with the point-intercept method.
Across datasets, the size of plots varied between 0.01 and
0.67 m2 (Table 1), but it was not correlated with latitude
(Spearman rs  20.05, p  0.86). Thus, although larger
plots were more effective in sampling rare species (significant positive correlation between plot size and number of
singletons: Spearman rs  0.79, p  0.0001) that are likely
to exhibit large temporal variability in cover, latitudinal patterns in stability and underlying mechanisms were independent of plot size.
In marine environments, where competition for primary
space among sessile organisms is generally strong, cover is an
important property of communities (Connolly and Muko
2003). In addition, since rocky benthic assemblages can be
multi-layered, the total percentage cover of organisms is not
constrained to 100% and can be reasonably taken as a proxy
for biomass (Cavanaugh et al. 2010).
For each plot, temporal stability of cover at the community level was estimated as the ratio between the mean
(across the time series) and the standard deviation of percentage cover (m/s, the inverse of the coefficient of variation;
Lehman and Tilman 2000). Since variations in temporal stability can be generated by changes in either the numerator,
the denominator or both, we formally analysed variations
in m and s, across a latitudinal gradient and following the
experimental removal of algal canopies.
The degree of synchrony of species fluctuations has
often been examined by calculating the sum of all pair-wise
species covariances in a plot and interpreting negative plot
covariances as evidence of the prevalence of asynchronous
fluctuations and, hence, of a positive contribution of the covariance effect to community stability (Lehman and Tilman
3
2000, Cottingham et al. 2001). The validity of this metric
has been recently challenged by Loreau and de Mazancourt
(2008), who have argued that it does not allow comparisons
between communities differing in species richness and that
the underlying assumption of species fluctuating independently is highly unrealistic. Here, community-wide synchrony
in species cover (ϕc ) was calculated for each plot by means
of a measure of synchrony recently proposed by the same
authors (Loreau and de2Mazancourt 2008). This is calculated
 s   , where s 2 is the variance in total
2
as ϕc  σcT
ci
cT
 i∑

1
community cover and sci is the standard deviation in cover
of species i, in a community with S species. This statistic
varies between 0 (perfect asynchrony) and 1 (perfect synchrony) and makes no specific assumption about the magnitude and distribution of species abundances and variances.
The strength of the statistical averaging effect is dependent upon the scaling relationship between the mean and
the variance (mean-variance rescaling, Tilman 1999). This
relationship follows Taylor’s power law: S 2  cmz, where S 2
is the variance, c is a constant, m is the mean and z is the
scaling coefficient. Larger values of z indicate a more rapid
increment in the variance with the mean and, assuming that
species in a community are independent, have equal abundances and mean-variance rescaling coefficients, statistical
averaging is supposed to promote stability whenever z  1.
Thus, z was calculated for each taxon in each plot as the
slope of the regression of log(variance) versus log(mean).
In order to assess how variations in the value of z could
be related to variations in species dominance (Doak et al.
1998, Cottingham et al. 2001), we assumed an exponential
rank–abundance distribution; taxa were ranked in decreasing
order of abundance, so that the contribution of the ith taxon
becomes mi  m1e2a(i21), where m1 is the most abundant
taxon and a is the coefficient determining the rate at which
mean abundance decreases with increasing rank (Doak et al.
1998). Thus, larger values of a indicate greater inequality in
abundance (i.e. lower evenness).
In general, if a species mixture performs better than component monocultures there is overyielding. However, as
noted by Fridley (2001), it is possible to distinguish between
the case in which a diverse assemblage performs more effectively than the average of the component monocultures (i.e.
richness effect sensu Cardinale et al. 2006) and that in which
a diverse assemblage performs more effectively then the best
performing component monoculture (i.e. complementarity
effect sensu Cardinale et al. 2006). Here, overyielding was
estimated as the slope of the regression of total community
cover versus taxa richness in each plot (Tilman 1999, Lehman
and Tilman 2000). Thus, following Cardinale et al. (2006),
it will hereafter be referred to as a richness effect. The different times of sampling were used as replicates and slopes were
calculated by means of a linear mixed effect model (Crawley
2002), in order to take into account the autocorrelation
in residuals (i.e. plots repeatedly sampled through time).
In order to correct for differences in the sampling effort
among descriptive datasets (the number of replicates varying between 5 and 8), weighted regression was used to
investigate the relationship between response variables and
latitude. Weights were calculated as the ratio between the
number of replicates available for each study and the variance
4
among them. Assumptions of linearity and homogeneity of
variances were checked by visual inspection of residual plots.
Log transformation was effective in removing non-linearity
and variance heterogeneity in response variables. BCa
(bias-corrected accelerated percentile) bootstrap confidence
intervals set at 95% were calculated using 999 iterations.
Unfortunately, the geographical positioning of sites for
which experimental data were available (Table 1) did not
allow assessment of variation in the effects of the removal
of macroalgal canopies along a latitudinal gradient using
a regression approach. Thus, data were analyzed by means
of a two-way factorial ANOVA, including the factors Site
(random) and Canopy (fixed, control versus removal). The
shores investigated were subsequently separated into northern and southern according to prevalent climatic conditions.
In particular, shores investigated along the coasts of Italy
and Portugal ( 45°N of latitude) are characterized by a
Mediterranean climate (Lusitanian-Mediterranean bioregion, with hot, dry summers and warm, wet winters), while
those investigated at higher latitudes ( 45°N of latitude),
except for those in the North Sea experiencing cold winters,
are under the influence of an oceanic climate (Eastern Atlantic Boreal bioregion, with mild winters and cool summers).
Evidence of variation in the effects of the canopy removal with
latitude was, however, not conclusive. Data for 8 out of the
9 sites analyzed had 5 replicates (n  5) for each of the control and the canopy removal treatment. In order to maintain
a balanced design and enable a straightforward analysis, the
same number of plots was randomly selected from the greater
number available for both control and canopy-removal plots
at the remainder site (i.e. n  8 experimental dataset 4, in
Table 1). Homogeneity of variances was checked by means
of Cochran’s test (Underwood 1997). In some cases, it could
not be achieved by transformation, but data were analyzed
nonetheless, since analysis of variance is robust to departure
from this assumption when there are many independent
replicates and sizes of samples are equal (Underwood 1997).
Results were, however, interpreted with caution, by judging
significance more conservatively (a  0.01). SNK tests were
used for a posteriori comparisons of the means.
Results
The stability of rocky shore assemblages decreased significantly from southern to northern shores (y  1.339 2 0.013x;
2
F1,94  8.31, p  0.01, RAdj
 0.07, CI  20.023, 20.005;
Fig. 1A). Both total assemblage cover (y  1.655  0.008x;
2
F1,94  10.22, p  0.01, RAdj
 0.09, CI  20.008, 0.017)
and standard deviation (y  0.348  0.021x; F1,94  17.25,
2
p  0.001, RAdj
 0.15, CI  0.004, 0.030) significantly
increased with latitude.
Species synchrony also increased with latitude
2
(y  22.608  0.034 x; F1,94  24.24, p  0.001, RAdj
 0.20,
CI  0.005, 0.049; Fig. 1B), indicating a weakening of
the covariance effect when moving from southern to northern rocky shores. In contrast, there was no significant relationship between latitude and richness effect (F1,94  3.25,
p  0.05), mean-variance scaling coefficient z (F1,94  1.94,
p  0.05) or evenness (F1,94  0.01, p  0.05). Importantly,
z, ranging between 1.19 and 1.79 across sites, was always
Temporal stability in cover (µ/σ)
(A)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Species synchrony (ϕ)
(B)
Temporal stability in cover (µ/σ)
40
45
40
45
50
55
50
Latitude
55
Latitude
0.0
–0.5
–1.0
–1.5
–2.0
–2.5
(C)
In general, the effects of the experimental removal of
canopy-forming macroalgae on stability and underlying
mechanisms varied among sites (Table 2). Removing the
canopy resulted in a significant decrease in temporal stability only at the four southernmost sites (Table 2, Fig. 2A).
Except for one site, it also caused a decline in mean total
cover (Table 2, Fig. 2B). Worth noting is that the magnitude
of this effect was particularly large at some of the northern
sites (Fig. 2B). In contrast, the standard deviation of cover
was significantly altered by the removal of the canopy only
at a few sites (Table 2); however, it tended to increase with
canopy loss at southern sites and decrease at northern sites
(Fig. 2C). Like standard deviation, significant effects of
the treatment on the synchronization of species fluctuations emerged throughout the latitudinal range investigated
(Table 2, Fig. 2D). However, species fluctuations tended
to become more synchronous at southern sites when algal
canopies were removed, while an opposite trend emerged for
sites at higher latitudes (Fig. 2D).
The removal of the canopy caused an increase in the
mean-variance scaling coefficient, z, suggesting a generalized enhancement of the statistical averaging; this effect
was, however, significant at some sites, but not at others
(Table 2, Fig. 2E). Likewise, removing canopies fostered
evenness, but only at some sites (Table 2, Fig. 2F). Finally,
the richness effect did not vary between control and canopy
removal plots (Table 2).
1.4
1.4
Discussion
1.2
1.0
0.8
0.6
0.4
0.2
0.0
–1.0
–0.5
0.0
0.5
Species synchrony (ϕ)
1.0
Figure 1. Regressions of (A) temporal stability in cover (m/s) and
(B) species synchrony (ϕ) against latitude. Estimates of temporal
stability and species synchrony have been log transformed before
the analysis to improve linearity and remove heterogeneity of variances. The relationship between stability and species synchrony (C)
is also reported to illustrate the importance of this mechanism in
determining variations in stability along a latitudinal gradient. The
use of residuals from the regression of species synchrony against
latitude guarantees that the relationship between stability and species synchrony was not generated by the correlation of both variables with latitude.
greater than 1, suggesting that the statistical averaging effect
was ubiquitous along the latitudinal range investigated. The
large decrease in temporal stability (m/s) with increasing synchronization (ϕc) of species fluctuations (y  0.697 2 0.317x;
2
F1,94  41.07, p  0.001, RAdj
 0.30, CI  20.402, 20.209;
Fig 1C) suggests that this mechanism was the main driver of
the reduction in stability towards higher latitudes.
The temporal stability (m/s) in the cover of rocky benthic communities tended to decrease with increasing latitude. This pattern can result from a decrease in the mean
or an increase in the standard deviation of the aggregate
property or both (Lehman and Tilman 2000). Both mean
total cover (m), and the standard deviation of cover (s),
increased with latitude, but the negative relationship
between stability (m/s) and latitude suggests that the proportional increment in total assemblage cover was smaller
than that of standard deviations. Statistical averaging
contributed to enhance stability (i.e. z  1), but effects
were invariant across the latitudinal gradient. Although
the relationship between cover and species richness (i.e.
richness effect) was positive (overyielding) in some plots
(∼ 78%) and negative (underyielding) in others (∼ 22%),
no general pattern emerged across the latitudinal range
investigated. Therefore, the reduction in stability towards
higher latitudes was mainly driven by greater synchronization of species fluctuations, ultimately resulting in positive
or less negative summed species covariances (Lehman and
Tilman 2000). These patterns indicate that compensatory
dynamics (i.e. the covariance effect) are neither ubiquitous nor rare, but they are likely to vary in space, being
influenced by a complex mix of biological and physical
processes (Valdivia and Molis 2009).
The prevalence of positive or less negative species covariances (i.e. synchronous species fluctuations) has often
been explained by inferring a greater signature of environmental forcing relative to that of biological processes, such
as competition or predation (Houlahan et al. 2007, Valone
5
Table 2. ANOVA of the effects of Treatment (canopy present vs canopy removal) and Site on temporal stability (m/s), mean cover (m), standard
deviation (s), species fluctuations (j), statistical averaging (z), richness effect and evenness (a). Degrees of freedom for each term and interaction included in the analysis are reported in brackets. **p  0.01; ***p  0.001.
Treatment  Tr (1)
MS
Site  Si (8)
F
Response variable
130.265
Temporal stability (m/s)
57470.630
Mean (m)
567.449
Std dev. (s)
0.176
Species synchrony (j)
Statistical averaging (z)
1.115
Richness effect
0.001
Evenness (a)
0.450
MS
30.37
5.03
4.95***
24.73** 9964.892 35.48***
0.73
1051.194
9.20***
3.07
0.083 10.21***
0.291 13.25***
20.67**
0.09
0.067 18.20***
1.95
0.355 13.76***
and Barber 2008). A recent meta-analysis, including both
aquatic and terrestrial environments, has reported a weakening of the stabilizing effect of richness as latitude increased
(Campbell et al. 2011). Speculating on this pattern, the same
(A)
Temporal
stability in cover (µ/σ)
16
*
12
8
*
*
*
4
F
MS
Transformation
Cochran’s
test
25.922
2324.129
775.277
0.058
0.054
0.007
0.231
4.22***
8.28***
6.79***
7.05***
2.46*
1.84
8.97***
6.136
280.833
114.261
0.008
0.022
0.004
0.026
none
none
none
none
none
none
ln(x  1)
p  0.01
p  0.05
p  0.01
p  0.05
p  0.01
p  0.01
p  0.05
authors (Campbell et al. 2011) have inferred that, at high
latitudes, environmental forcing would prevail on biological
drivers (i.e. negative species interactions, such as competition
and predation) in shaping the relationship between diversity
(B)
250
200
0
Species synchrony (ϕ)
Standard
deviation in cover (σ)
(D)
*
80
60
40
*
*
*
20
0
*
*
*
100
*
*
*
50
1.0
0.8
0.6
*
0.4
*
*
0.2
*
*
*
0.0
2.5
(F)
1.5
1.0
2.0
*
*
1.5
*
*
*
*
1.0
0.5
0.0
*
*
150
Evenness (a)
Statistical averaging (z)
(E)
Control
Removal
0
100
(C)
Residual (72)
Tr  Si (8)
MS
F
Mean total cover (%)
Source of variation
1
2
3
Southern Europe
4
5
6
7
8
Northern Europe
Site
*
0.5
0.0
9
*
1
2
3
Southern Europe
4
5
6
7
8
9
Northern Europe
Figure 2. Effects of canopy removal at different European sites on (A) stability in community cover (m/s), (B) mean community cover (m),
(C) standard deviation in community cover (s), (D) species synchrony (ϕ), (E) mean-variance scaling coefficient (z) and (F) evenness (a).
Sites on the x-axis are ordered following an increasing latitudinal gradient according to the numbering presented in Table 1 and divided in
the groups Southern Europe (below 45°N) and Northern Europe (above 45°N). Data are means  SE (n  5). Asterisks indicate significant
differences between the control and the canopy removal treatment for each site.
6
and stability. The same mechanism could be invoked as an
explanation for the greater synchrony of species fluctuations
on rocky shores at higher latitudes, where fluctuations in
physical variables (e.g. light, temperature) are wider than on
shores at lower latitudes.
Stress due to desiccation and heat is an important determinant of the structure and dynamics of rocky intertidal
assemblages (Connell 1972). As shown by Helmuth et al.
(2002), thermal stress is regulated by the interplay between
tidal amplitude and timing of low tides and, hence, it does
not decrease linearly with increasing latitude. Indeed, on
the west coast of the United States, some northern sites
were more thermally stressful than some southern sites
for intertidal invertebrates, suggesting that the interaction
between climate and timing of low tides can create a complex mosaic of thermal environments along a gradient of
latitude (Helmuth et al. 2002). Detailed information on
tidal regimes, comparable to that produced by Helmuth
et al. (2002) for the west coast of the United States, is, to
the best of our knowledge, not available for the European
coasts investigated. Nonetheless, if we assume the occurrence of a mosaic of thermal environments (Helmuth et al.
2002), the linearity in the variation of stability and species
synchrony with latitude would suggest a greater influence
of other environmental variables covarying with latitude
(i.e. light, temperature) than that of tidal amplitude and
timing of low tides. This does not, by any means, imply
that the interaction between the tidal regime and local
climate plays no role in generating latitudinal patterns in
stability and species synchrony at the sites we investigated.
Alternatively, variation in temporal dynamics of populations might underlie the latitudinal pattern in species synchrony. The temporal turn-over of species composition slows
down towards higher latitudes, as a consequence of lower
mean temperatures (Shurin et al. 2007). The temporal scale
necessary to observe strong diversity effects on stability might
be longer at high latitudes, due to slower population dynamics of species (Campbell et al. 2011). Thus, when investigated over a two-year temporal scale, the stabilizing effect
of compensatory dynamics could have emerged on shores at
lower latitudes, where species turn-over is faster, but not on
those at higher latitudes, where species turn-over is slower.
The removal of canopy-forming macroalgae had a strong
negative impact on temporal stability only at four sites;
interestingly, these were the southernmost investigated.
Removing algal canopies caused a generalized decrease in mean
covers, but this was smaller at the four most southern sites
(ranging between 20.0 and 30.0%; mean  SE: 21  3.8)
than at the five most northern sites (ranging between 40.0
and 97.0%; mean  SE: 71.85  10.6). The standard deviation of cover was significantly altered by the removal of the
canopy only at a few sites, but it tended to increase with
canopy loss at southern sites and decrease at northern sites
(Fig. 2C).
Thus, at southern sites, changes in both m (decreasing)
and s (increasing) of cover after canopy removal were destabilizing in direction and caused a decline in temporal stability. At these sites, the drop in mean community cover
caused by the canopy removal was not as large as it could
be expected from the large ratio of canopy cover to that of
understorey assemblages (Table 1). This suggests that understorey species compensated (i.e. increased in cover) for the
decline in mean community cover caused by the elimination
of the canopy-forming species. A trend for species fluctuations to become more synchronous following the removal of
the canopy would confirm that most species responded positively to changed environmental conditions. Thus, at Sites
2 and 3, the enhanced strength of the statistical averaging
effect caused by the removal of the canopy was not sufficient
to counter destabilizing effects of smaller community cover
and more positive species covariances.
At northern sites (except for Site 4), the large decrease in
mean community cover (m), mainly generated by the loss
of the canopy-forming species, was compensated for by a
decrease in the standard deviation of community cover (s).
This ultimately resulted in little overall change in temporal
stability (m/s). Similar compensation between destabilizing effects arising from decreased abundance and stabilizing effects, due to decreased temporal variability, have been
recently documented in a subtidal marine system (Long
et al. 2011).
Which mechanisms sustained community stability at
northern sites, in the face of destabilizing effects of decreased
mean cover? The removal of the canopy, generating less
synchronous species fluctuations, promoted the stabilizing
effects of compensatory dynamics. This, along with a generalized strengthening of statistical averaging, was effective in
countering destabilizing effects due to the decline in community cover (Fig. 2). Thus, although the loss of dominant
canopy-forming species led to a decline of community cover,
the strengthening of the mechanisms promoting stability
in response to disturbance was able to provide an insurance
against the collapse of this ecosystem function. This suggests that a large abundance of particularly stable species (i.e.
canopy-forming species on rocky shores) does not always
enhance overall community stability (Lepš 2004, Polley et al.
2007, Grman et al. 2010).
The occurrence of different canopy-forming species
across the European shores investigated prevents assessing to what extent the impact of their loss on stability and
underlying mechanisms was driven by local environmental
conditions or by species-specific life traits. On southern
shores, manipulated canopy-forming species varied from
Chondrus crispus and Mastocarpus stellatus on Atlantic coasts
of Portugal (Site 2), to smaller sized Cystoseira compressa
and C. amentacea var. strycta on shores located in SE
(Site 1) and NW (Site 3) Italy, respectively. Although these
species largely differ in both architecture and size, their
removal caused a consistent increase in the synchron­
ization of species fluctuations and a drop in stability on
southern shores. In contrast, on northern shores, the effects
of removing Fucus spp. were markedly different from those
of removing the canopies formed by Ascophyllum nodosum
(at Site 4) that were more similar to those recorded at
southern sites.
The ratio between canopy and understorey assemblage
covers might have played an important role in determining the response of species fluctuations and stability to the
removal of canopies formed by different species. This ratio
was generally small at sites were canopies where made by
7
Fucoids, while it was larger in the presence of other canopyforming species (i.e. at Sites 1 to 4). This suggests that
species-specific traits of habitat-formers could generate
variation in their effects on the synchronization of species
fluctuation and stability of associated assemblages. More
specifically, attributes of Fucoids, such as size, architecture
and toughness of tissues, might modify environmental
conditions in such a way to enhance the synchronization of
species fluctuations.
On the other hand, effects of canopy-forming macro­
algae on understorey organisms have been previously shown
to vary with latitude (Leonard 2000). For instance, on the
east coast of the USA, the effects of canopies formed by
Ascophyllum nodosum on associated organisms (i.e. barnacles) were positive at southern sites, but negative at northern
ones (Leonard 2000). In addition, even slight variations
in climate (i.e. maximum daily temperatures) were sufficient
to alter the sign of the interactions between the habitatformer and the associated species (Leonard 2000). More
generally, the occurrence of positive species interactions is
predicted to increase along an increasing gradient of stress
(i.e. the stress-gradient hypothesis; Bertness and Callaway
1994). While predictions from this conceptual framework
have been tested and refined in a variety of environments
(Callaway 2007, Brooker et al. 2008, Bulleri 2009), the
implications of shifts in the sign and strength of species
interactions on stability and underlying mechanisms remain
largely unexplored.
By dampening variations in physical conditions, the
presence of macroalgal canopies would be predicted to
reduce the signature of environmental forcing, resulting
in the prevalence of biological interactions, such as competition and, hence, of less synchronous species fluctuations (Houlahan et al. 2007, Valone and Barber 2008). On
shores at higher latitudes, species fluctuations showed, in
contrast, a consistent tendency to become less synchronous
following the experimental removal of the canopy-forming
Fucus spp. This means that most species started fluctuating
more asynchronously when fully exposed to variations in
physical variables (e.g. light and temperature) that are particularly wide at high latitudes. Thus, on northern shores,
the effects of biological forcing (i.e. the presence of canopyforming macroalgae) in synchronising species fluctuations
overwhelmed those of environmental forcing.
At the four most southern European sites, synchrony in
species fluctuations tended to increase in canopy removal
plots. Amelioration of harsh physical conditions (i.e. desiccation and heat) by the canopy could have enhanced negative
species interactions among understorey species, in line with
predictions from the stress-gradient hypothesis (Bertness
and Callaway 1994). Once deprived of a canopy, most of
the species could have tracked more closely variations in
environmental conditions, resulting in a greater synchron­
ization of their fluctuations (Houlahan et al. 2007, Valone
and Barber 2008, Engelen and Santos 2009).
Our findings, together with those of Campbell et al.
(2011), point towards the occurrence of a latitudinal trend in
the stability of natural communities. This would be mostly
produced by variations in the contribution of compensatory
dynamics, at odds with the more consistent contribution
of statistical averaging and richness effect. In addition, our
8
results suggest that the potential of biological interactions
to generate positive covariances (i.e. biological forcing)
has been greatly overlooked (but see Wilby and Shachak
2004, Steiner et al. 2005). In some cases, habitat-formers
would be able to synchronize species fluctuations over the
levels generated by local physical conditions. Thus, greater
synchronization of species fluctuations does not necessarily
reflect a prevalence of environmental forcing over biological
forcing, as often assumed (Houlahan et al. 2007, Valone
and Barber 2008, Campbell et al. 2011). For these reasons,
framing the biodiversity–stability relationship into facilitation
theories, such as the stress-gradient hypothesis (Bertness and
Callaway 1994), might contribute to reconcile contrasting
effects of biodiversity reported in the literature (Cottingham
et al. 2001) by enabling a more balanced account of abiotic
and biotic forces at play.
Acknowledgements – We sincerely thank D. Vasseur and P. Petraitis
for providing valuable comments on an earlier version of the ms.
The final version has been greatly improved by insightful comments
and constructive criticism from M. Bracken. The project has been
carried out in the framework of the MarBEF Network of Excellence
‘Marine Biodiversity and Ecosystem Functioning’ which was
funded by the Sustainable Development, Global Change and
Ecosystems Programme of the European Community’s Sixth
Framework Programme (contract no. GOCE-CT-2003-505446).
While writing, FB was supported from an Assegno di Ricerca
from the University of Sassari.
References
Airoldi, L. and Beck, M. W. 2007. Loss, status and trends for
coastal marine habitats of Europe. – Oceanogr. Mar. Biol.
Annu. Rev. 45: 345–405.
Benedetti-Cecchi, L. et al. 2001. Predicting the consequences of
anthropogenic disturbance: large-scale effects of loss of canopy
algae on rocky shores. – Mar. Ecol. Prog. Ser. 214: 137–150.
Bertness, M. D. and Callaway, R. 1994. Positive interactions in
communities. – Trends Ecol. Evol. 9: 191–193.
Bertness, M. D. and Ewanchuk, P. J. 2002. Latitudinal and
climate-driven variation in the strength and nature of biological interactions in New England salt marshes. – Oecologia
132: 392–401.
Brooker, R. W. et al. 2008. Facilitation in plant communities:
the past, the present and the future. – J. Ecol. 96: 18–34.
Bruno, J. F. and Bertness, M. D. 2001. Habitat modification
and facilitation in benthic marine communities. – In: Bertness
M. D. et al. (eds), Marine community ecology. Sinauer,
pp. 201–218.
Bulleri, F. 2009. Facilitation research in marine systems: state of the
art, emerging patterns and insights for future developments.
– J. Ecol. 97: 1121–1130.
Callaway, R. M. 2007. Positive interactions and interdependence
in plant communities. – Springer.
Campbell, V. et al. 2011. Experimental design and the outcome
and interpretation of diversity-stability relations. – Oikos 120:
399–408.
Cardinale, B. J. et al. 2006. Effects of biodiversity on the functioning
of trophic groups and ecosystems. – Nature 443: 989–992.
Cavanaugh, K. C. et al. 2010. Scaling giant kelp field measurements to regional scales using satellite observations. – Mar.
Ecol. Prog. Ser. 403: 13–27.
Chapin, F. S. et al. 1998. Ecosystem consequences of changing
biodiversity - experimental evidence and a research agenda for
the future. – Bioscience 48: 45–52.
Coleman, R. A. et al. 2006. A continental scale evaluation of
the role of limpet grazing on rocky shores. – Oecologia 147:
556–564.
Connell, J. H. 1972. The ecology of rocky shores. – Annu. Rev.
Ecol. Syst. 3: 169–192.
Connolly, S. R. and Muko, S. 2003. Space preemption, sizedependent competition, and the coexistence of clonal growth
forms. – Ecology 84: 2979–2988.
Connolly, S. R. et al. 2001. A latitudinal gradient in recruitment
of invertebrates in the northeast Pacific Ocean. – Ecology 82:
1789–1813.
Cottingham, K. L. et al. 2001. Biodiversity may regulate the temporal variability of ecological systems. – Ecol. Lett. 4: 72–85.
Crawley, M. J. 2002. Statistical computing: an introduction to
data analysis using S-Plus. – Wiley.
Doak, D. F. et al. 1998. The statistical inevitability of stability–
diversity relationships in community ecology. – Am. Nat. 151:
264–276.
Downing, A. L. et al. 2008. Environmental fluctuations induce
scale-dependent compensation and increase stability in plankton ecosystems. – Ecology 89: 3204–3214.
Engelen, A. and Santos, R. 2009. Which demographic traits
determine population growth in the invasive brown seaweed
Sargassum muticum? – J. Ecol. 97: 675–684.
Fischer, J. M. et al. 2001. Compensatory dynamics in zooplankton
community responses to acidification: measurement and
mechanisms. – Ecol Appl. 11: 1060–1072.
Fridley, J. D. 2001. The influence of species diversity on ecosystem
productivity: how, where and why? – Oikos 93: 514–526.
Griffin, J. N. et al. 2009. Biodiversity and the stability of ecosystem
functioning. – In: S. Naeem, D. E. et al. (eds), Biodiversity,
ecosystem functioning, and human wellbeing: an ecological
and economic perspective. Oxford Univ. Press, pp. 78–93.
Grman, E. et al. 2010. Mechanisms contributing to stability in
ecosystem function depend on the environmental context.
– Ecol Lett. 13: 1400–1410.
Helmuth, B. et al. 2002. Climate change and latitudinal patterns
of intertidal thermal stress. – Science 298: 1015–1017.
Houlahan, J. E. et al. 2007. Compensatory dynamics are rare in
natural ecological communities. – Proc. Natl Acad. Sci. USA
104: 3273–3277.
Huston, M. A. 1997. Hidden treatments in ecological experiments: re-evaluating the ecosystem function of biodiversity.
– Oecologia 110: 449–460.
Isbell, F. I. et al. 2009. Biodiversity, productivity and the temporal
stability of productivity: patterns and processes. – Ecol. Lett.
12: 443–451.
Ives, A. R. et al. 1999. Stability and variability in competitive
communities. – Science 286: 542–544.
Ives, A. R. et al. 2000. Stability and species richness in complex
communities. – Ecol. Lett. 3: 399–411.
Jenkins, S. R. et al. 1999. Direct and indirect effects of a macroalgal
canopy and limpet grazing in structuring a sheltered inter-tidal
community. – Mar. Ecol. Prog. Ser. 188: 81–92.
Keitt, T. H. 2008. Coherent ecological dynamics induced by
large-scale disturbance. – Nature 454: 331–334.
Klug, J. L. et al. 2000. Compensatory dynamics in planktonic
community responses to pH perturbations. – Ecology 81:
387–398.
Kraufvelin, P. et al. 2006. Nutrient addition to experimental rocky
shore communities revisited: delayed responses, rapid recovery.
– Ecosystems 7: 1076–1093.
Lehman, C. L. and Tilman, D. 2000. Biodiversity, stability and
productivity in competitive communities. – Am. Nat. 156:
534–552.
Leonard, G. H. 2000. Latitudinal variation in species interactions:
a test in the New England rocky intertidal zone. – Ecology 81:
1015–1030.
Lepš, J. 2004. Variability in population and community biomass
in a grassland community affected by environmental productivity and diversity. – Oikos 107: 64–71.
Li, W. and Stevens, M. H. H. 2010. How enrichment, ecosystem
size, and their effects on species richness co-determine the stability of microcosm communities. – Oikos 119: 686–695.
Lilley, S. A. and Schiel, D. R. 2006. Community effects following
the deletion of a habitat-forming alga from rocky marine
shores. – Oecologia 148: 672–681.
Long, Z. T. et al. 2011. Food chain length and omnivory determine
the stability of a marine subtidal food web. – J. Anim. Ecol.
80: 586–594.
Loreau, M. and de Mazancourt, C. 2008. Species synchrony and
its drivers: neutral and nonneutral community dynamics in
fluctuating environments. – Am. Nat. 172: 48–66.
McCann, K. S. 2000. The diversity–stability debate. – Nature 405:
228–233.
Pennings, S. C. and Silliman, B. R. 2005. Linking biogeography
and community ecology: latitudinal variation in plant–herbivore
interaction strength. – Ecology 86: 2310–2319.
Polley, H. W. et al. 2007. Dominant species constrain effects of
species diversity on temporal variability in biomass production
of tallgrass prairie. – Oikos 116: 2044–2052.
Romanuk, T. N. and Kolasa, J. 2002. Environmental variability
alters the relationship between richness and variability of
community abundances in aquatic rock pool microcosms.
– Ecoscience 9: 55–62.
Romanuk, T. N. et al. 2006. Nutrient enrichment weakens the
stabilizing effect of species richness. – Oikos 114: 291–302
Sanford, E. et al. 2003. Local selection and latitudinal variation
in a marine predator–prey interaction. – Science 300:
1135–1137.
Shurin, J. B. et al. 2007. Diversity–stability relationship varies with
latitude in zooplankton. – Ecol. Lett. 10: 127–134.
Steiner, C. F. et al. 2005. The influence of consumer diversity
and indirect facilitation on trophic level biomass and stability.
– Oikos 110: 556–566.
Thom, R. M. and Widdowson, T. B. 1978. A resurvey of E.
Yale Dawson’s 42 intertidal algal transects on the southern
California mainland after 15 years. – Bull. S. Calif. Acad. Sci.
77: 1–13.
Tilman, D. 1996. Biodiversity: population versus ecosystem
stability. – Ecology 77: 350–363.
Tilman, D. 1999. The ecological consequences of changes in
biodiversity: a search for general principles. – Ecology 80:
1455–1474.
Underwood, A. J. 1997. Experiments in ecology. – Cambridge
Univ. Press.
Valdivia, N. and Molis, M. 2009. Observational evidence of
a negative biodiversity-stability relationship in intertidal
epibenthic communities. – Aquat. Biol. 4: 263–271.
Valone, T. J. and Barber, N. A. 2008. An empirical evaluation
of the insurance hypothesis in diversity–stability models.
– Ecology 89: 522–531.
Vasseur, D. A. and Gaedke, U. 2007. Spectral analysis unmasks
synchronous and compensatory dynamics in plankton communities. – Ecology 88: 2058–2071.
Vermeij, G. J. 1978. Biogeography and adaptation. – Harvard
Univ. Press.
Vogt, H. and Schramm, W. 1991. Conspicuous decline of Fucus
in Kiel Bay (western Baltic): what are the causes? – Mar. Ecol.
Prog. Ser. 69: 189–194.
Wilby, A. and Shachak, M. 2004. Shrubs, granivores and annual
plant community stability in an arid ecosystem. – Oikos 106:
209–216.
Yachi, S. and Loreau, M. 1999. Biodiversity and ecosystem
productivity in a fluctuating environment: the insurance
hypothesis. – Proc. Natl Acad. Sci. USA 96: 1463–1468.
9