ArticlePDF Available

A general biodiversity-function relationship is mediated by trophic level

June 2016
Oikos 126(1)

DOI:10.1111/oik.03652

Authors:

Mary I. O’Connor

University of British Columbia

Andrew Gonzalez

McGill University

Jarrett Byrnes

University of Massachusetts Boston

Show all 13 authorsHide

Species diversity affects the functioning of ecosystems, including the efficiency by which communities capture limited resources, produce biomass, recycle and retain biologically essential nutrients. These ecological functions ultimately support the ecosystem services upon which humanity depends. Despite hundreds of experimental tests of the effect of biodiversity on ecosystem function (BEF), it remains unclear whether diversity effects are sufficiently general that we can use a single relationship to quantitatively predict how changes in species richness alter an ecosystem function across trophic levels, ecosystems and ecological conditions. Our objective here is to determine whether a general relationship exists between biodiversity and standing biomass. We used hierarchical mixed effects models, based on a power function between species richness and biomass production (Y = a × Sb), and a database of 374 published experiments to estimate the BEF relationship (the change in biomass with the addition of species), and its associated uncertainty, in the context of environmental factors. We found that the mean relationship (b = 0.26, 95% CI: 0.16, 0.37) characterized the vast majority of observations, was robust to differences in experimental design, and was independent of the range of species richness levels considered. However, the richness–biomass relationship varied by trophic level and among ecosystems; b was nearly twice as large for consumers (herbivores and detritivores) compared to primary producers in aquatic systems; in terrestrial ecosystems, b for detritivores was negative but depended on few studies. We estimated changes in biomass expected for a range of changes in species richness, highlighting that species loss has greater implications than species gains, skewing a distribution of biomass change relative to observed species richness change. When biomass provides a good proxy for processes that underpin ecosystem services, this relationship could be used as a step in modeling the production of ecosystem services and their dependence on biodiversity. This article is protected by copyright. All rights reserved.

Content uploaded by John Griffin

Content may be subject to copyright.

A general biodiversity–function relationship is mediated by trophic level

Mary I. O’Connor1, Andrew Gonzalez2, Jarrett E. K. Byrnes3, Bradley J. Cardinale4, J. Emmett Duffy5, Lars

Gamfeldt6, John N. Griffin7, David Hooper8, Bruce A. Hungate9, Alain Paquette10, Patrick L. Thompson2,

Laura E. Dee11 and Kristin L. Dolan12

1Dept of Zoology and Biodiversity Research Centre, Univ. of British Columbia, Vancouver, BC, V6T 1Z4,

Canada

2Dept of Biology, McGill Univ., Montreal, QC Canada

3Dept of Biology, Univ. of Massachusetts Boston, Boston, MA, USA

4School of Natural Resources and Environment, Univ. of Michigan, Ann Arbor, MI, USA

5Tennenbaum Marine Observatories Network, Smithsonian Institution, Washington DC, USA

6Dept of Biological and Environmental Sciences, Univ. of Gothenburg, Göteborg, Sweden

7Dept of Biosciences, Swansea Univ., Singleton Park, Swansea, UK

8Dept of Biology, Western Washington Univ., Bellingham, WA, USA

9Center for Ecosystem Science and Society, Dept of Biological Sciences, Northern Arizona Univ., Flagstaff,

AZ, USA

10Centre for Forest Research, Univ. du Québec à Montréal, Centre-ville Station, Montréal, QC, Canada

11Inst. on the Environment, Univ. of Minnesota, Twin Cities, Saint Paul, MN, USA.

12Research Development Office, Univ. of California at San Francisco, San Francisco, CA, USA

Corresponding author: M. I. O’Connor, Dept of Zoology and Biodiversity Research Centre, Univ. of British

Columbia, Vancouver, BC, V6T 1Z4, Canada. E-mail: oconnor@zoology.ubc.ca

Decision date: 13-Jun-2016

This article has been accepted for publication and undergone full peer review but has not

been through the copyediting, typesetting, pagination and proofreading process, which may

lead to differences between this version and the Version of Record. Please cite this article

as doi: [10.1111/oik.03652].

Accepted Article

(Abstract)

Species diversity affects the functioning of ecosystems, including the efficiency by which communities

capture limited resources, produce biomass, recycle and retain biologically essential nutrients. These

ecological functions ultimately support the ecosystem services upon which humanity depends. Despite

hundreds of experimental tests of the effect of biodiversity on ecosystem function (BEF), it remains unclear

whether diversity effects are sufficiently general that we can use a single relationship to quantitatively predict

how changes in species richness alter an ecosystem function across trophic levels, ecosystems and ecological

conditions. Our objective here is to determine whether a general relationship exists between biodiversity and

standing biomass. We used hierarchical mixed effects models, based on a power function between species

richness and biomass production (Y = a × Sb), and a database of 374 published experiments to estimate the

BEF relationship (the change in biomass with the addition of species), and its associated uncertainty, in the

context of environmental factors. We found that the mean relationship (b = 0.26, 95% CI: 0.16, 0.37)

characterized the vast majority of observations, was robust to differences in experimental design, and was

independent of the range of species richness levels considered. However, the richness–biomass relationship

varied by trophic level and among ecosystems; b was nearly twice as large for consumers (herbivores and

detritivores) compared to primary producers in aquatic systems; in terrestrial ecosystems, b for detritivores

was negative but depended on few studies. We estimated changes in biomass expected for a range of changes

in species richness, highlighting that species loss has greater implications than species gains, skewing a

distribution of biomass change relative to observed species richness change. When biomass provides a good

proxy for processes that underpin ecosystem services, this relationship could be used as a step in modeling the

production of ecosystem services and their dependence on biodiversity.

Accepted Article

Introduction

A major goal in biodiversity research is to understand the consequences of biodiversity change for ecosystem

functioning (Tilman et al. 1997, Petchey 2000, Turnbull et al. 2013). Experiments have shown that species

richness positively affects many ecosystem functions, such as standing biomass and resource use (Tilman et

al. 2001, Cardinale et al. 2006, Reich et al. 2012). A nonlinear function captures the relationship between

species richness and ecosystem functions, and its prevalence among experimental results suggests a common

quantitative relationship might characterize the rate of change of function with changing species richness.

Generalized empirical relationships in ecology have allowed for comparisons and predictions across complex

systems (Peters 1983, Brown and West 2000). Here we tested whether a general empirical relationship

adequately describes the relationship between diversity and biomass production, assessing the degree to

which this relationship depends on both random experimental factors and a variety of ecological parameters

(e.g., ecosystem type and trophic level). We use this variation to estimate changes in biomass expected with

changes in species richness of different organism types.

To estimate how biodiversity change will influence changes in ecosystem functioning in contexts

beyond controlled experiments, we need (1) a quantitative estimate of how much function is lost with the loss

of a species, (2) reliable estimates of variation around the mean estimate of the BEF relationship and, ideally

(3) assignment of uncertainty to factors that are known to influence the relationship (e.g., species traits,

resource supply, ecosystem type) as well as factors not yet identified. With existing data, an empirical

estimate of the relationship between richness and biomass could be used in biodiversity change models to

give a first approximation, or testable prediction, for effects of biodiversity change outside experimental

settings. The ecosystem function of biomass production, here estimated as standing biomass at a particular

time point and referred to throughout as ‘standing biomass’, has often been described as a positive

decelerating function of species richness (e.g., Balvanera et al. 2006, Cardinale et al. 2006, Reich et al. 2012).

For competitively structured communities, this relationship may follow the Michaelis-Menten function;

however in many experiments, the saturation of biomass production with accumulating species is not clear at

the levels of species richness tested (Cardinale et al. 2011). An alternative model that captures the strong

effects of species richness at low levels of richness, but diminishing effects at higher richness, is a power

function, biomass = a*(richness)b in which b describes the relationship between a change in richness and

Accepted Article

biomass (when 0 < b < 1) (Cardinale et al. 2007, 2011, Reich et al. 2012). The power function used here is not

meant to imply a particular theoretical BEF mechanism. It is used because it has substantial empirical support

from previous syntheses (Cardinale et al 2011, Gamfeldt et al. 2015, Lefcheck et al 2015).

To what extent might the response of biomass production to species richness be generalizable across

ecosystems and ecological contexts? If the BEF relationship varies in space and time, or with biotic and

abiotic conditions, then estimates of b alone may impart minimal information for understanding the role of

biodiversity. However, if it is general or varies predictably, the value of b can provide a powerful tool for

efforts to generalize the consequences of species loss for ecological function and ecosystem services (Isbell et

al 2014). Previous grassland experiments reported a central tendency toward a value of approximately b =

0.26 for effects of species richness on biomass, but with substantial variation in this estimate. Ninety-five

percent confidence intervals ranged from 0.15 - 0.32 (Cardinale et al. 2006), or a standard deviation of 0.27

(Cardinale et al. 2011). Whether that variation reflects systematic and ecologically important differences

among BEF relationships in different systems remains an important question.

The evidence is mixed as to whether abiotic and biotic conditions influence the value of b. Individual

experimental studies suggest that the change in biomass with accumulating species richness can vary with

resource availability (e.g., water, nutrients, CO2) (Reich et al. 2001, Fridley 2002, Boyer et al. 2009) or

presence of a predator (Duffy et al. 2005), and can increase in strength over time (Stachowicz et al. 2008,

Reich et al. 2012). Within experiments that share a species pool, experimental design, and other factors, the

BEF relationship can also vary among sites (Hector et al. 1999). Such among-site variation could imply that

the strength of the richness-function relationship is contingent on species composition and local

environmental condition (e.g., soil fertility, climate, etc)(Hooper et al. 2012). In contrast, meta-analyses of

dozens of experiments have demonstrated that, across studies, estimated mean richness-standing biomass

relationships (e.g., b) or effect sizes (e.g., log response ratios) are conserved across experiments conducted in

different ecosystem types and trophic groups (Cardinale et al. 2006). However, the values of b do vary

systematically with attributes of the experimental design, such as additive and substitutive designs or the total

number of richness levels (Balvanera et al. 2006), experimental durations (Cardinale et al. 2007), and spatial

and temporal scale (Cardinale et al. 2011). In addition, four recent meta-analyses reported differences in BEF

effects among trophic levels: marine herbivore richness had a stronger effect on function than richness of

Accepted Article

algae (Gamfeldt et al. 2014), aquatic herbivore richness had stronger effects on function than plant richness

when multiple functions are analyzed (Lefcheck et al. 2015), carnivore richness more strongly affected

resource depletion than did richness at lower trophic levels (Griffin et al. 2013), and detritivore richness more

strongly increased decomposition rates than did plant litter richness (Srivastava et al. 2009, Hooper et al.

2012). While individual experiments, and broader meta-analyses, have tested the importance of one or a few

additional factors (time, resource supply, trophic structure, etc; (Hooper et al 2012, Tilman et al 2012), the

relative importance of these factors, the uncertainty in their effects, and whether it is necessary to include

these parameters in general richness-function models remains unclear.

Here, we test the hypothesis that a single BEF relationship, expressed as an empirically estimated

value of b in a power function, adequately describes the relationship between species richness and standing

biomass at fine spatial grains (e.g., m2 or litres) despite variation across experiments in abiotic conditions,

sites, and ecological communities. We then tested biological and experimental conditions, such as different

species pools, ecosystem types, trophic levels, resource regimes, and lengths and types of experiments, that

might explain variation in this relationship, aiming to identify which factors are essential to understanding the

richness-biomass relationship and which might be left out of a general model. We applied our findings to

estimate the effects of changes in species richness for changes in standing biomass. Ultimately, our goal is to

facilitate integration and quantitative application of the BEF relationship by determining whether

experimental evidence supports a general, quantitative relationship (a general b value) between richness and

the important ecological function of community biomass production.

2. Methods

We used a hierarchical mixed effects model to test our hypothesis that there is a constant relationship

between species richness and community biomass. We chose standing biomass as the response variable,

because theoretical work has centered on this response and hundreds of experimental tests of the relationship

between community biomass and richness are published. Standing biomass is closely correlated with net

primary production under certain conditions, such as when biomass turnover or size structure is is constant

across treatments, or per capita production rates scale isometrically with body size at the individual and

population level. When these conditions are not met, biomass may not approximate productivity rates. The

Accepted Article

data most available - and analyzed here - are for standing stocks only. Modeling standing biomass provides a

diverse and large sample to estimate not only the effect of richness on standing biomass but also to test for

systematic variation in this relationship among groups.

Our sample of studies included 374 experiments and 558 entries (from 91 studies published between

1985 and 2009, Appendices A and B) in which richness was manipulated and standing stock of biomass was

reported for a species assemblage. The fundamental unit of observation in our analysis was a biomass

response variable (e.g., above-ground biomass, density, etc) reported across a set of species richness (S)

treatments varying in the number of species (at least 2 richness levels), with all other factors controlled,

hereafter an “entry.” For most entries, we lacked data on individual replicates for a treatment in a given

experimental unit, and had no choice but to use published means of the richness treatment. Though all entries

shared this basic experimental design, they differed in 1) the number of richness levels tested, 2) maximum

species richness, 3) the duration of the experiment, 4) whether resources like nutrients or water were added,

reduced or unmanipulated, 5) whether experiments were conducted in the lab or in the field, 6) whether the

ecosystem studied was aquatic or terrestrial, and 7) in which trophic level diversity was manipulated and

biomass reported (Table 1). Many experiments included monocultures (richness = 1 species), such that 544 of

558 entries included S = 1. Across all experiments, the highest richness level tested (Smax) increased with the

number of richness levels within entries (r2 = 0.41, P < 0.001). For each entry, we obtained or estimated a

value for each predictor listed above (Table 1). Studies were dropped from the analysis when information on

this set of predictors was not available, so there were no unknown values, and the dataset included the same

information for all models tested. The number of entries for each level of each predictor was not balanced.

The database is dominated by terrestrial plant studies lacking explicit resource manipulations (Table 1). The

database also does not include the following combinations: aquatic species richness x resource reduction

treatments, terrestrial herbivore richness manipulations, or resource reduction treatments for herbivores or

detritivores. There were also insufficient studies reporting the effects of carnivore diversity on carnivore

biomass to include in this analysis. Fortunately, hierarchical mixed effects models handle unbalanced designs,

and groups with few data points can still contribute some information to the overall analysis (Gelman and Hill

2007).

Accepted Article

2.1 The search for a single BEF relationship

Our primary objective was to estimate the relationship between richness and biomass. Then, we

aimed to test whether a single b coefficient described the relationship between richness and biomass given the

variation across organisms, ecosystems, and studies performed to date, and if not, to determine what

additional information is required to estimate the effect of species richness on biomass. We chose a mixed

effects modeling approach that allowed us to characterize the effect of richness on biomass using our

structured dataset in which many variables are shared by observations reported from the same experiment or

study.

In this dataset, entries within experiments differ in aspects including date sampled or response

variable (e.g., above or below ground biomass sampled from the same plot), but share all other attributes such

as species richness levels, focal taxa, etc. Experiments within studies differ in treatment levels of resources,

location or time (e.g., year sampled), but share a publication, research team, and other study-level attributes

(Table 1). Mixed effects models allow modeling of variation associated with all unmeasured variables that

make parameter estimates from the same group (e.g., a study) similar to each other but distinct from other

groups. Hierarchical mixed effects models pool information at the group level, using fewer degrees of

freedom and reducing uncertainty in estimated relationships relative to an analysis of each group (e.g., study)

independently with regressions (Pinheiro and Bates 2000, Gelman and Hill 2007, O’Connor et al. 2007,

Cressie et al. 2009). Hierarchical mixed effects models that account for such structure in datasets are used

extensively in social sciences, economics, public health, and other fields where grouped data are the norm

(Snijders and Bosker 1999, Gelman and Hill 2007), and provide an information-efficient approach for

structured data. Hierarchical modeling is an established ecological research tool well-suited to large datasets

comprised of similar, smaller datasets (Myers and Worm 2003, O’Connor et al. 2007, Bolker et al. 2009,

Hudson et al. 2013, Lefcheck et al. 2015)

We modeled the biomass-richness relationship at the finest data resolution with the simplest plausible

relationship of interest, derived by natural-log-transforming the power function Y = a*Sb. In our case biomass

(ln(Y)) at the plot level as predicted by species richness (ln(S)),

ln(Yijkl) = Β0.ijk + Β1.ijk*ln(Sijkl) + εijkl, Eqn 1a

Accepted Article

so that a is estimated by Β0 and the parameter b is estimated by Β1 for plots or mesocosms (l) within each

combination of a species richness manipulation and a biomass response (entry, k), entries within experiments

(j), and experiments within studies (i). We assumed normally distributed residual error (εijk ~ (N, σ2)).

Although other formulations have been used to describe this relationship (e.g. Michaelis-Menten, Cardinale et

al., 2011), we proceed with the power function, which has also received substantial empirical support and

offers greater analytical simplicity, though differs from Michaelis-Menten in that it does not saturate

(Cardinale et al. 2007, 2011, Reich et al. 2012, Gamfeldt et al. 2014).

Our hypotheses are centered on the question of how predictable is the value of Β1, the slope of

biomass on species richness, or conversely, how variable it is among studies and conditions. Though it is not

of primary interest in this study, we also modeled variation in the intercept term, Β0, because predictors of Β1

could influence the intercept (mean biomass), and those influences likely co-vary in some cases with variation

in the slope. The intercept term in a BEF regression model represents the absolute value of biomass at

standard richness level. Biomass varies among groups of organisms for many reasons – taxonomy of the

group involved (algae vs grass vs insects), absolute resource supply rates, etc. Our analysis and dataset are not

suited to modeling biomass variation among experimental units (the intercept term). Although we model the

intercept, because it may explain some variation in slopes as discussed above, we do not interpret variation in

the intercept estimate in terms of the predictors we have included, because we know this set of predictors is

insufficient for understanding variation in biomass (the intercept) among entries, experiments and studies.

To test our hypotheses while accounting for variability among experimental conditions and study

systems in our dataset, we modeled variation in the slope (Β1.ijk) and intercept (Β0.ijk):

Β0.ijk = γ00 + μ0.i + μ0.j + μ0.k Eqn 1b

Β1.ijk = γ10 + μ1.i + μ1.j + μ1.k

In Eqns 1b, the slope Β1.ijk and intercept Β0.ijk for each observation (a set of species richness – biomass

observations) are modeled as mean γ10 and γ00, respectively. Variation associated with each level of data

grouping - entry (μ1.k, μ0.k), experiment (μ1.j, μ0.j) and study (μ1.i, μ0.i) - can be formally considered as random

effects normally distributed with variance Σ0 estimated by the model (Appendix C).

The test of our first hypothesis, that there is a constant relationship between species richness and

standing biomass, is whether variable slopes (b, as estimated by Β1.ijk) are required among different studies,

Accepted Article

experiments and entries (μ1.i, μ1.j, or μ1.k ≠ 0, in Eqn 1b) (Table 2). If so, we would conclude that it is not

possible to identify a single parameter (b) for this dataset. We also tested alternate models (Eqns 2-3) that

include interaction terms for time (TG) passed since the beginning of the experiment normalized to the

generation time of the taxon (such that TG = duration of experiment in days / generation time of focal

organism) (Β2), and the interaction between TG and ln(S) (Β3):

ln(Yijkl) = Β0.ijk + Β1.ijk*ln(Sijkl) + Β2*ln(TG.ijkl) + εijkl. Eqn 2

ln(Yijkl) = Β0.ijk + Β1.ijk*ln(Sijkl) + Β2*ln(TG.ijkl) + Β3*ln(Sijkl)*ln(TG.ijkl) + εijkl. Eqn 3

These models test for effects of plot-scale richness and plot age, and are possible because paired richness,

function data were reported for multiple time points in many studies. Generation time was approximated

based on body size and knowledge of taxa (Cardinale et al, 2011).

2.2 Testing hypotheses about factors that modify the BEF relationship

We tested our second main hypothesis that ecological or experimental parameters that varied across

entries, experiments or studies altered the richness-biomass relationship (Table 2). Specifically, we compared

mixed effects models with different formulations that represent hypotheses for how various biotic and abiotic

factors (listed in Table 1) interact with species richness to affect the relationship.

In addition to the basic hypothesis that biomass changes with increasing species richness and time

(models 1-3), we tested the hypothesis that ecosystem (aquatic, terrestrial) and trophic group (primary

producer, detritivore, herbivore) influence the richness-biomass relationship (slope = Β1.ij, model 4, Tables 2,

4). The trophic group predictor indicates the group for which species richness was manipulated and biomass

was measured. In this hypothesis, we included an interaction between ecosystem and trophic group to allow

for the lack of data on terrestrial herbivores. We also tested the hypotheses that in addition to ecosystem and

trophic group, increased or reduced resources (water, nutrients, CO2) modified the BEF relationship (model

5). The three categorical levels of the resource treatment predictor (control, addition, reduction) reflect

experimental manipulations relative to ambient conditions for any resource explicitly manipulated (water,

nitrogen, light, etc). A level of ‘control’ was assigned to any species richness manipulation that did not

specify that resources were added or reduced relative to ambient levels. Some experiments manipulated

resource supply to plants and factorially with consumer richness manipulations, and we included these

Accepted Article

studies. This resource predictor includes no information on whether the resource was a priori shown to be

limiting in the system, and not all experiments included factorial resource treatments (Table 1). Consequently,

the ‘resource’ predictor represents a coarse test of whether resource manipulation modifies the richness-

function relationship.

In a fourth hypothesis, we tested for an effect of experimental duration among studies standardized

to estimated generation time of the manipulated taxa, testing whether experiments that run for a greater

number of generations show stronger effects of richness when compared across taxa or systems (model 6,

Tables 2, 4). We considered time in two ways. First, we modeled the effect of time ‘within an experiment’,

looking at whether the slope parameter changes as an experiment moves from year 1 to year 2 to year 3. We

might expect the parameter to change over time based on studies in long-term experiments such as Reich et al

2012 and Stachowicz et al 2008. This effect of time is captured by the parameter TG, and models the effect of

year (or day) within a multi-year (day) experiment (Table 2, model 2). Second, we examined the effect of

time by modeling the effect of total experiment duration on the slope b. This model tested whether longer

experiments have steeper slopes (the parameter is called ‘ln(maxDuration)’, which is measured by the number

of generations of the focal taxa in the experiment) (Table 2, model 6). We also tested whether the effect of

total experiment duration depended on whether resources were added or reduced (Table 2, model 7).

We also tested the hypothesis that the BEF relationship varies with attributes of the experimental

design – maximum duration, maximum number of species tested (Smax), units in which biomass was measured

(biomass estimator), and lab vs field (model 8). Finally, we tested the hypothesis that all factors modify the

BEF relationship (model 9, Eqn 4), and that when all are included, the interaction between ecosystem and

trophic group is not important (model 9.1). We modeled interactions between intercepts (Β0.ijk) and slopes

(Β1.ijk) for each group using the following equations, and each hypothesis outlined above was modeled as a

nested subset of the full model:

Β0.ijk = γ00 + γ01*Sysi + γ02*Li + γ03*Sysi*Li + γ04 *Unitsj + γ05*LabFieldi

+ γ06*Smax.il + γ07* Nj + γ08*ln(max(Durationi)) + μ0.i + μ0.j + μ0.l Eqn 4a

Β1.ijk = γ10 + γ11*Sysi + γ12*Li + γ13*Sysi*Li + γ14 *Unitsj + γ15*LabFieldi

+ γ16*Smax.il + γ17* Nj + γ18*ln(max(Durationi)) + μ1.i + μ1.j + μ1.k Eqn 4b

Accepted Article

with random effects, normally distributed about zero with variance estimated by the model (Appendix C).

2.3 Model selection, analysis and inference

To identify the best model, we first determined the need for variable slopes and intercepts for each

candidate model (Eqn 1a, 2 and 3) by comparing models with different random effects structures (Table A1).

The test of our first hypothesis is whether the BEF model requires variable slopes at group (entry, experiment

and study) levels, implying variation in the BEF relationship among groups. We ranked models with and

without variable slopes and intercepts using AICc adjusted for degrees of freedom to account for different

random effects following Bolker et al. (2009) and Gelman and Hill (2007), and compared them using δAICc

values (Bolker et al. 2009). If variable slopes were required at the group level, we examined residuals (μ1.k,

μ1.j, μ1.l) using caterpillar plots to determine whether only a few studies drove the need for variable slopes at

the group level (Verbeke and Molenberghs 2000).

To test our second hypothesis, we compared models with biotic and experimental predictors (models

4-9; Eqns 4a-b). We ranked models using AICc, and compared them with δAICc and Akaike weights (w). We

defined the best model set as all models with δAICc < 2, (Richards 2005, Burnham and Anderson 2002). If

more than one model met our criteria of δaic < 2, we averaged these models to produce coefficient estimates

(Burnham and Anderson 2002). Model averaging produces estimates for all coefficients in the best model set,

weighted by the importance (w) of each model in the set. To estimate the parameter b for each study, we

summed coefficients for each richness manipulation (b = Β1.ijk + μ1.i + μ1.j + μ1.k) (Gelman and Hill 2007) from

the best model set.

We proceeded with a linear mixed effects model, although in our dataset, ln(Yijkl) values have a fat-

tailed distribution and are not strictly normally distributed (Shapiro Test, p < 0.001), differing from normal

but without significant skew. Analysis of residuals of Equation 1 revealed 7 experiments from 2 studies that

were extreme outliers in the dataset (< 3% of entries), and these were excluded from analysis to meet

assumptions of homoscedasticity. Although we tested for an effect of time and there is a risk that observations

are temporally autocorrelated, we could not include a temporal autocorrelation term in the model because

time and richness are modeled at the finest resolution of our hierarchical data. Thus, there are multiple

observations (biomass at multiple richness levels) for each level of TG within each entry, and we cannot

Accepted Article

isolate potential autocorrelation in time from among richness levels. Therefore, the ln(TG)*ln(S) fixed effect is

expected to include any real effects and any possible autocorrelation. All analyses were done in R (v. 3.2.1).

We used lme4 package (v. 1.1-8) for mixed effects analyses, comparing models fit with REML = FALSE but

used REML = TRUE for estimation of coefficients. Data and analytical code are available from the authors at

https://github.com/mioconnor78/OConnor-et-al-BEF-Relationship. We produced caterpillar plots using the

package sjPlot (v. 1.8.2).

To consider the relationship between a change in richness and a change in function, we simply

generated a normal distribution of numbers between 0 and 2, centered on 1, to represent a distribution of

proportional species richness changes centered on no net change. We then used the empirical estimates of b

produced from the previous analyses to translate a distribution in species richness change to an expected

distribution of change in biomass. We present these as proportional changes in species richness and

proportional changes in function to compare the effect of species loss or gain across systems with different

total species richness.

3. Results

A single, universal relationship (Β1.ijk value) was not supported by our analysis of the species richness-

biomass relationship. Variable slopes and intercepts associated with entry, experiment and study were

required for each candidate model (Eqns 1-3) (AICC > 100 for comparison of model with variable slopes and

intercepts to model with fewer random effects terms) (Table A1). We did not find strong evidence for a

systematic effect of experimental duration on the BEF relationship across all studies (Table 3). This simple

model with random effects (model 2; Eqn 2) estimates a BEF relationship of b = γ10 = 0.23 (95% CI: 0.18,

0.28) that applies to most (> 94.00%) but not all entries (Figure A1, Table A1). Examination of the variation

in slopes (μ1.i, μ1.j and μ0.k), plotted as the deviation of each slope’s estimated random effect from the mean

slope fixed effect (Figure 2), suggests this estimate of b = γ10 adequately described most observations (i.e., the

confidence intervals for the random effects include 0 in the caterpillar plots for most ln(S) estimates) (Figure

A1). Still, the number of slope residuals deviating from the central estimate (γ10) is sufficient that removing

those observations neither eliminates the need for variable slopes, nor is justified based on the dataset. In all

models, richness values were centered on the value that minimized co-variances of random effects for slopes

Accepted Article

and intercepts estimated by model 1 (ln(8). These covariances were 0.07 for Entry, 0.24 for Experiment, and -

0.16 for Study. Examination of standardized residual plots suggests models were not overfit (Figure A3).

After concluding that variable slopes at the entry, experiment and study levels confound the

identification of a single relationship, we tested our second main set of hypotheses that ecological and

experimental factors could explain some variation in richness-biomass relationship, thus eliminating the need

for variable slopes (μ1.i, μ1.j and μ0.k) based on diagnosis using caterpillar plots (Figures A1, 2). We found that

the BEF relationship varied systematically between aquatic primary producers and consumers (Model 4,

Table 4) such that herbivore biomass increased with species richness by baq.herbivores = γ10 + γ12 = 0.47, whereas

detritivore biomass increased with species richness by baq. detritivores = γ10 + γ12 = 0.55, both stronger than the

relationship between primary producer (plants, algae) biomass and primary producer species richness

(bprim.prod = γ10 = 0.26) (Table 4, Figure 1B). Though there was no difference between aquatic and terrestrial

primary producers, terrestrial detritivores had a much weaker relationship between richness and biomass than

all other groups (aquatic primary producers and herbivores, and terrestrial plants), with a negative value for b

(Table 5, Figure 1B, C). Estimates for the intercept term varied among trophic and ecosystem groups, as

expected by their very different biomasses (Table 5, Figure A2). The top-ranked model of our set was model

4 (Table 4), which included the interaction between trophic group and ecosystem (Figure 1B). None of our

other hypotheses about variation in the BEF relationship were comparable to this ‘best’ model (AIC weight =

0.784).

The best model indicates that variable slopes and intercepts are still required, even with fixed effects

for trophic group and ecosystem (Model 4, Table A2, Figure 2). Thus, systematic variation remains among

entries, experiments, and studies that prohibits a single estimate of a BEF relationship between ln(Y) and

ln(S) (Figure 2). The larger variance components associated with study and entry compared to experiment

suggests that most of the unexplained systematic variation is at those levels. Slope estimates did not differ

systematically for experiments with or without monocultures (Figure A4).

Our model comparison results allowed us to reject some of our alternate hypotheses (Table 4). We

rejected the hypothesis that the basic model (Eqn 2) is sufficient to explain the relationship between richness

and biomass. We also rejected the hypotheses (models 8, 9 and 9.1) that differences in experimental designs

(number of species tested, lab vs field experiment, and the method of estimating biomass) explain variation in

Accepted Article

the observed BEF relationship, and that our coarse grouping of resource supply condition (addition, reduction

or control) provided needed information for understanding the richness – biomass relationship. Results of the

test of the importance of number of generations (TG) within experiments did not support the hypothesis that

including the interaction between ln(S) and ln(TG) led to a significantly better fit (Table 3). AICc values and

the likelihood ratio test suggest that the richness-biomass relationship did not depend on the duration of an

experiment expressed as the generation time of the organisms being studied (e.g., no. generations, L) across

the 374 experiments in our dataset (no support for the interaction between time and richness, model 2 in Table

4). We conducted two additional tests of the hypothesis that experimental duration might affect the strength of

the relationship. In the first of these, we expanded model 3 (Table 4) to test alternate hypotheses that there is

an interaction between ecosystem, time and richness (Model 3a) or between trophic group, time and richness

(Model 3b)(Table A4). The inclusion of the interaction term for time in these models suggests an effect of

time could be informative, yet coefficients for the TG interactions did not differ from 0 except for herbivores,

which suggests a weak negative effect of time on the BEF relationship. Second, we tested the effect of

maximum duration on the relationship for only final observations of each experiment (models 6, 7). Model

comparisons for this dataset were consistent with the full dataset, and suggested no effect of maximum

duration on the parameter b (Table A3). We did not find evidence for this effect of time on plants or

detritivores across all studies (Tables S4 and S5).

When applied to a distribution of scenarios of species richness change over time, the estimated

values of b (Table 5) produced distributions of expected biomass change that reveal net negative effects of

species richness change on biomass. A distribution of species richness changes centered on no change (or

proportion of richness before:after change = 1), produces a distribution of expected biomass change with a

mean proportional change <1 (Figure 3), and greater extreme values for loss of function than for gain. The

larger b-value for aquatic herbivores suggests much greater losses or gains in biomass expected for a given

change in species richness relative to primary producers. For example, a 20% loss of species richness for

herbivores leads to a 10% loss in herbivore biomass, while the same loss of plant species richness leads to a

6% loss of plant biomass.

4. Discussion

Accepted Article

We synthesized over 500 experimental tests of the effect of species richness on biomass, and found strong

support for a relationship that varies among trophic levels (Figure 1B-C) and potentially across ecosystem

types. We found that most (but not all) primary producer assemblages in both terrestrial and aquatic

environments conform to a single coefficient (b = 0.26) quantifying how biomass increases with increasing

species richness. However, aquatic consumers had much stronger effects of diversity on biomass (herbivores,

b = 0.47; detritivores, b = 0.54), and terrestrial detritivores had no clear relationship (b = -0.001), compared to

plants and algae. We conclude that information about trophic group and ecosystem can inform estimates of

the consequences of species loss or gain for one ecosystem function, standing biomass. Our hierarchical

mixed effects modeling approach provided one of the more comprehensive analyses of the richness-biomass

relationship to date, simultaneously considering the potential dependence of the richness-biomass relationship

on 8 abiotic and biotic factors and additional systematic variation across hundreds of experimental tests.

Finally, our results indicate that, on average, species losses will result in greater losses of biomass than will

species gains result in increased biomass. We discuss the potentially important implications of this result later

in the Discussion.

The observed stronger BEF relationship in primary consumers relative to primary producers has been

predicted conceptually (Duffy 2002, 2003). Although early data syntheses did not detect this difference, that

could be explained by smaller numbers of studies and relatively simple statistical methods of data synthesis

(Cardinale et al. 2006). Recently, using variants of the dataset we used here, Gamfeldt et al (2014) found that

in marine studies, herbivore biomass increases more strongly with richness than does primary producer

biomass, and Lefcheck et al. (2015) reported stronger effects of aquatic herbivore than primary producer

diversity on multiple functions. Similarly, Griffin et al (2013) found stronger effects of species richness on

resource depletion rates for higher trophic groups. We confirm this result for a larger dataset that includes

terrestrial studies, suggesting that as more data has become available, previous findings that herbivores did

not differ from plants can now be revised.

The larger effects of species richness change for aquatic herbivore and detritivore biomass than

primary producer biomass leads to the hypothesis that changes in diversity could create positive feedbacks in

aquatic systems. Because the magnitudes of the consequences increase nonlinearly as species richness

declines, greater diversity declines among consumers than resources (Byrnes et al. 2007, Duffy 2003) could

Accepted Article

shift the importance of top down control: as grazer species loss disproportionately reduces grazer biomass,

plant assemblages released from grazing pressure could increase plant productivity. Similar mechanisms

could influence the relationship between species richness and biomass for decomposers, and also for

decomposition rates. A meta-analyses based on the same data set found that changes in detritivore diversity

had greater effects on changes in decomposition rates than did changes in plant litter diversity, particularly

among aquatic detritivores (Hooper et al. 2012).

We found that patterns in the richness-biomass relationship across trophic levels differed in

terrestrial systems. In this dataset, detritivores were represented by primary decomposers: unicellular fungi

(Replanksy and Bell 2009), multicellular fungi (Setälä and McLean 2004), and bacteria (Griffiths et al. 2001)

(3 studies, 24 Experiments with distinct species compositions at each richness level, 24 Entries). While the

former two studies found evidence for positive effects of fungal diversity on fungal biomass without

disturbance, disturbance by drought reversed this effect (Setälä and McLean 2004), and bacterial diversity had

no effect on a variety of soil processes (Griffiths et al. 2001). The pattern of weaker richness-biomass

relationships among terrestrial detritivores relative to primary producers in this small set may reflect the

taxonomic bias toward microbial consumers in terrestrial systems relative to larger-bodied detritivores (e.g.,

macroinvertebrates) in the aquatic studies included here. However, the interpretation of a general, cross-

system trophic level effect in terrestrial systems is hampered by lack of available data. The few studies using

terrestrial detritivores are insufficient to understand the generality or causes of variation in that relationship.

Furthermore, our dataset included no terrestrial herbivore manipulations for which herbivore biomass was

reported, so we have no basis for inference about that relationship. Similarly, we lacked sufficient estimates of

carnivore biodiversity manipulations that reported effects on carnivore biomass to include them in this

analysis. These issues highlight the need for further exploration of biodiversity-ecosystem functioning

relationships in these understudied groups.

We did not find a systematic relationship between the BEF relationship and experimental duration

(TG) across this dataset (Table 3). Still, for several reasons, we cannot reject the hypothesis that the BEF

relationship changes though time within a community. First, there is strong evidence in the literature,

including one meta-analysis, that have reported that the richness-biomass relationship strengthens through

time (Cardinale et al. 2007, Stachowicz et al. 2008, Reich et al. 2012). Further, in some of the longest-running

Accepted Article

BEF experiments, an effect of duration is clear after several years (Stachowicz et al. 2008, Reich et al. 2012).

Second, time may have had variable effects among studies. We found that entry-level random effects for the

coefficient b were required, and one of the main differences between entries within an experiment is the time

of measurement. The persistence of the entry-level random effect for the BEF relationship could reflect

temporal variation, that variation may not be linear through time in all studies, or that co-varying factors such

as climate conditions explain effects associated with time. A similar argument could be made for the

importance of total experimental duration (maximum duration, Table 1), which varies among studies and

could therefore also be accounted for in the study-level variance component. At the study-level, variation in

study duration is typically confounded with variation in spatial scale and body size of the focal taxa

(Cardinale et al 2011), such that time effects cannot be clearly distinguished.

Our failure to reject the need for variable slopes among certain groupings (entries, experiments, and

studies) indicates that systematic variation in the BEF relationship exists among studies and experiments, not

captured by our hypotheses. This suggest that additional research and synthesis is needed to determine

whether there is a single BEF relationship, or whether attributions such as climate or higher resolution

treatment of predictors, such as resource supply, could explain the remaining variation. Among-study

variation explained the majority of the variation in the random effects in our model (Table 5). Random, study-

level variation is distinguishable from residual variation (error) and implies that in addition to the fixed effects

that we modeled, there is still systematic variation in how richness affects function among studies. This

variation could result from climate, site environmental parameters (e.g., soil pH), taxonomic groups studied,

species or functional trait composition within those groups, or other ecological or scientific particularities of

the research studies.

Our results also help shed light on potential changes in function resulting from changes in local

species richness. Our results show that, on average, species losses lead to greater losses of biomass than

species gains lead to increased biomass (Figure 3). The disproportionate effects of species losses compared to

gains follows from Jensen’s inequality theorem, which shows that the change in y per increase in x is different

than the change in y per the same unit decrease in x for any convex or concave function (Figure A4). Our

analyses confirmed that, despite some systematic variation, the relationship between species richness and

biomass is almost always concave down with positive values of b for the power function 0 < b < 1. Thus, on

Accepted Article

average, species gains accrue less change in biomass than is lost when an equivalent number of species are

lost.

There are several implications of Jensen’s inequality that warrant consideration in future studies.

First, Jensen’s inequality suggests that efforts to conserve species could have disproportionately large effects

on ecological functions compared to efforts to restore a similar number of species that have already been lost

from a system. Indeed, restoration of ecosystem functions may require the addition of a larger number of

native species than were originally lost. Jensen’s inequality may also be important for assessing arguments in

an ongoing debate about how local changes in species richness impact ecosystem functioning (Gonzalez et al.

In Press, Vellend et al. 2013). Vellend et al. (2013) recently summarized time-series from 346 studies that

had monitored plant species richness in ‘local-scale’ vegetation plots. These authors found that species

richness has increased through time in roughly half the plots, but decreased through time in the other half.

When averaged together, Vellend et al. concluded there has been no ‘net’ change in terrestrial plant species

richness, and went on to argue that, if there has been no net change in species richness, then there cannot be

changes in ecological function driven by local species loss.

Gonzalez et al. (in press) have criticized the study of Vellend on multiple grounds: (1) their dataset

was not representative of global patterns of plant species richness nor the primary drivers of diversity change,

(2) their dataset was unduly influenced by studies performed in ecosystems where biodiversity is likely

recovering from historical destruction (e.g., forests recovering from logging), and (3) the unjustified logic of

overextending conclusions of monitoring programs of biodiversity to ecological functions that were never

measured in the original studies. Our results add a fourth issue to consider when interpreting the functional

implications of changes in biodiversity in such syntheses as Vellend et al 2013. Even if Vellend et al.’s

primary conclusion that species gains and losses have averaged out to no ‘net’ loss of species richness is

correct, it is still incorrect to suggest that no net loss of richness means there has been no net loss of

ecological function. Assuming that species being gained and lost are, on average, functionally similar, our

results suggest that losses could still have disproportionately large impacts on productivity compared to

additions.

Strengths and limitations of the empirical relationship

Accepted Article

The main insight supported by our analysis is that there is empirical evidence to support the use of a

single value of b (in a power function) to describe how a change in species richness leads to a change in

biomass for primary producers, but distinct values for aquatic herbivores and detritivores. The strength of this

approach is a large database of experimental observations using similar experimental designs and a range of

taxa (Table 1). Our analysis does not provide information about the other parameter in the power function, a,

which can be thought of as the intercept of a linearized power function. Conceptually, a is the biomass of the

average monoculture for a given community, and therefore will vary substantially among communities

depending on their traits and environment. For example, a will be very low for phytoplankton and quite high

for shrub assemblages, and with these community types will vary with resource availability, climate, etc. We

did not have estimates for predictor variables that would be suitable to model variation in a among the very

different study systems and environments included in this synthesis. For any specific application of the

empirically estimated scaling exponent we provide here, the value of a will need to be estimated for the

ecosystem under consideration.

The implications of our finding that a single value of b applies to most observations for application

to other trophic groups or to support inferences about theoretical mechanisms remain limited for two reasons.

First, some predictors in our analysis should be interpreted with caution. For example, studies differed widely

in whether and how resources were controlled or manipulated. Thus, our predictor of ‘resource level’ is

coarse and does not represent resource limitation in these systems. Resource manipulation (addition, control

or reduction) was included in a plausible (but unlikely) model (Table 3). Based on the model ranking and the

coarseness of the biological meaning of the resource predictor, we do not reject the hypothesis that resource

supply can change the BEF relationship. Our analysis was limited by sufficient resource limitation data to

conclusively test this hypothesis. Previous studies have shown mixed results, with some individual studies

finding that increased nitrogen availability led to greater diversity effects on aboveground production (Reich

et al. 2001, Fridley 2003) and a meta-analysis, using some of the same data as our study finding the opposite

(Hooper et al. 2012). Some of this variability, and that found in our current study, could result from different

effects on aboveground versus belowground versus total production, effects of different resources (e.g., CO2

versus nutrients), different levels of resource addition, and compositional variation among communities

(Reich et al. 2001, Fridley 2002, Hooper et al. 2012). More work is needed to fully test the dependence of the

Accepted Article

BEF relationship on resource supply, ideally with studies that factorially manipulate both diversity and

resource supply within expected ranges of environmental change. Even with the heterogeneity in the data and

the coarseness of some predictor levels, the patterns we observe are consistent with previous findings from

analyses that tested a subset of these predictors on smaller datasets.

Finally, while not the objective of our study, the existence of a power law scaling relationship in

other areas of ecology has been inferred to imply self-similar systems and a certain class of mechanistic driver

(Brown et al 2002). We caution that our empirical study, fitting a power function to BEF data, does not imply

such a mechanistic driver to a BEF relationship. Determining whether such a relationship exists would merit

further theoretical development, including assessing whether a power law is indeed the best descriptor of the

BEF relationship. Instead, we aimed to test for a general empirical pattern. It remains to be determined

whether there is a single best functional form to describe the BEF relationship, and whether this relationship

is predicted or explained by any single theoretical framework.

Conclusion

Our analysis of the richness-biomass relationship allows practitioners to apply an empirically-

derived, a priori prediction for the BEF relationship as a quantitative estimate for the expected importance of

a change in biomass with a change in species richness. This estimate provides a starting hypothesis that

investigators can use to determine whether additional factors modify the diversity-biomass relationship, or

that they can attempt to falsify or improve upon. Furthermore, when biomass provides a good proxy for the

processes and functions that underpin ecosystem services, this estimate of b could be used as a step in

modeling the production of ecosystem services and their dependency on biodiversity. For

instance, this BEF relationship can be part of an ecosystem service production function

(Barbier 2007, Isbell et al. 2014), where production functions describe the relationship

between various inputs (e.g., ecosystem properties, harvesting effort, etc.) and the level of a

service that is produced (Barbier 2007). These production functions can support

management decisions targeting provisioning of ecosystem services, such as by evaluating

ecosystem service provisioning under different scenarios (Barbier 2007, Nelson et al. 2009,

Accepted Article

Tallis and Polasky 2009). Such an approach can also determine how different estimates of

this parameter influence estimates of ecosystem service supply, and provide insight into the

marginal value of maintaining diversity in terms of the value of an ecosystem service.

However, for many ecosystem services, standing biomass is not a direct proxy for many ecosystem

services that directly contribute to human well-being (for example, secondary productivity or nutrient

cycling). Therefore, there is a need to determine whether this relationship holds more generally for other

response variables that are also closely linked to human well-being (e.g., food production, water quality), and

to what extent these findings extend to cases of non-random species loss. If so, integrating such a relationship

into production functions could represent an important step towards the development of new tools to forecast

the magnitude of change in important ecosystem services due to biodiversity loss, for a broader array of

services. In the meantime, there is sufficient evidence to support the application of this parameterized power

function to efforts such as integrated ecosystem function models or the generation of production functions

linking biodiversity change to ecosystem functions and services directly related to biomass.

Acknowledgements:

This work was conducted as a part of the Biodiversity and the Functioning of Ecosystems: Translating Model

Experiments into Functional Reality Working Group supported by the National Center for Ecological

Analysis and Synthesis, a Center funded by NSF (Grant #EF-0553768), the University of California, Santa

Barbara, and the State of California. MO is supported by an NSERC discovery grant and the Alfred P. Sloan

Foundation. AG is supported by an NSERC discovery grant, the Canada Research Chair Program and the

Quebec Centre for Biodiversity Science.

Authorship statement, following ICMJE guidelines:

MO and AG designed this work and drafted the manuscript, MO conducted the analysis, BC and KD acquired

the data, all authors contributed to interpretation of the analysis and implications, revision of intellectually

important content, and approved of its final publication and have agreed to be accountable for the work.

Accepted Article

References:

Balvanera, P. et al. 2006. Quantifying the evidence for biodiversity effects on ecosystem functioning and

services. - Ecol. Lett. 9: 1–11.

Barbier, E. B. 2007. Valuing ecosystem services as productive inputs. - In: in press.

Bolker, B. M. et al. 2009. Generalized linear mixed models: a practical guide for ecology and evolution. -

Trends Ecol. Evol. 24: 127–135.

Boyer, K. E. et al. 2009. Biodiversity effects on productivity and stability of marine macroalgal

communities: the role of environmental context. - Oikos 118: 1062–1072.

Brown, J. H. and G. West. 2000. Scaling in Biology (JH Brown and GB West, Eds.). - Oxford University

Press.

Burnham, K. P. and Anderson, D. R. 2002. Model selection and multimodel inference. - Springer-Verlag.

Cardinale, B. J. et al. 2006. Effects of biodiversity on the functioning of trophic groups and ecosystems. -

Nature 443: 989–992.

Cardinale, B. J. et al. 2007. Impacts of plant diversity on biomass production increase through time

because of species complementarity. - Proc. Natl. Acad. Sci. U. S. A. 104: 18123–18128.

Cardinale, B. J. et al. 2011. The functional role of producer diversity in ecosystems. - Am. J. Bot. 98: 572–

592.

Cressie, N. et al. 2009. Accounting for uncertainty in ecological analysis: the strengths and limitations of

hierarchical statistical modeling. - Ecol. Appl. 19: 553–570.

Duffy, J. E. 2002. Biodiversity and ecosystem function: the consumer connection. - Oikos 99: 201–219.

Duffy, J. E. 2003. Biodiversity loss, trophic skew and ecosystem functioning. - Ecol. Lett. 6: 680–687.

Duffy, J. E. et al. 2005. Ecosystem consequences of diversity depend on food chain length in estuarine

vegetation. - Ecol. Lett. 8: 301–309.

Fridley, J. D. 2002. Resource availability dominates and alters the relationship between species

diversity and ecosystem productivity in experimental plant communities. - Oecologia 132:

271–277.

Fridley, J. D. 2003. Diversity effects on production in different light and fertility environments: an

experiment with communities of annual plants. - J. Ecol. 91: 396–406.

Gamfeldt, L. et al. 2014. Effects of changes in species richness in the ocean - a meta-analysis. 124: 252–

265.

Gelman, A. and Hill, J. 2007. Data analysis using regression and multilevel/hierarchical models. -

Cambridge University Press.

Gonzalez, A. et al. In Press. Estimating local biodiversity change: a critique of papers claiming no net

loss in local diversity. - Ecology in press.

Griffin, J. N. et al. 2013. Effects of predator richness on prey suppression: a meta-analysis. - Ecology 94:

2180–2187.

Accepted Article

Hector, A. et al. 1999. Plant Diversity and Productivity Experiments in European Grasslands. - Science

286: 1123–1127.

Hooper, D. U. et al. 2012. A global synthesis reveals biodiversity loss as a major driver of ecosystem

change. - Nature 486: 105–U129.

Hudson, L. N. et al. 2013. The relationship between body mass and field metabolic rate among

individual birds and mammals. - J. Anim. Ecol. 82: 1009–1020.

Isbell, F. et al. 2014. The biodiversity-dependent ecosystem service debt. - Ecol. Lett. in press.

Lefcheck, J. S. et al. 2015. Biodiversity enhances ecosystem multifunctionality across trophic levels and

habitats. - Nat. Commun. in press.

Myers, R. A. and Worm, B. 2003. Rapid worldwide depletion of predatory fish communities. - Nature

423: 280–283.

Nelson, E. et al. 2009. Modeling multiple ecosystem services, biodiversity conservation, commodity

production, and tradeoffs at landscape scales. - Front. Ecol. Environ. 7: 4–11.

O’Connor, M. I. et al. 2007. Temperature control of larval dispersal and implications for marine ecology,

evolution, and conservation. - Proc. Natl. Acad. Sci. 104: 1266–1271.

Petchey, O. L. 2000. Species diversity, species extinction and ecosystem function. - Am. Nat. 155: 696–

702.

Peters, R. H. 1983. The ecological implications of body size. - Cambridge University Press.

Pinheiro, J. and Bates, D. 2000. Mixed-effects models in S and S-PLUS. - Springer-Verlag.

Reich, P. B. et al. 2001. Plant diversity enhances ecosystem responses to elevated CO2 and nitrogen

deposition. - Nature 410: 809–812.

Reich, P. B. et al. 2012. Impacts of biodiversity loss escalate through time as redundancy fades. - Science

336: 589–592.

Snijders, T. A. B. and Bosker, R. J. 1999. Multilevel Analysis: An Introduction to Basic and Advanced

Multilevel Modeling. - Sage Publishers.

Srivastava, D. S. et al. 2009. Diversity has stronger top-down than bottom-up effects on decomposition. -

Ecology 90: 1073–1083.

Stachowicz, J. J. et al. 2008. Diversity enhances cover and stability of seaweed assemblages: the role of

heterogeneity and time. - Ecology 89: 3008–3019.

Tallis, H. and Polasky, S. 2009. Mapping and valuing ecosystem services as an approach for

conservation and natural-resource management. - In: The Year in Ecology and Conservation

Biology 2009. in press.

Tilman, D. et al. 1997. Plant diversity and ecosystem productivity: Theoretical considerations. - Proc.

Natl. Acad. Sci. 94: 1857–1861.

Tilman, D. et al. 2001. Diversity and Productivity in a Long-term Grassland Experiment. - Science 294:

843–845.

Accepted Article

Turnbull, L. A. et al. 2013. Coexistence, niches and biodiversity effects on ecosystem functioning (JHR

Lambers, Ed.). - Ecol. Lett. 16: 116–127.

Vellend, M. et al. 2013. Global meta-analysis reveals no net change in local-scale plant biodiversity over

time. - Proc. Natl. Acad. Sci. 110: 19456–19459.

Verbeke, G. and Molenberghs, G. 2000. Linear mixed models for longitudinal data. - Springer-Verlag.

Accepted Article

Table Legends

Table 1. Summary of hierarchical dataset on the richness – biomass relationship and predictors analyzed in

this study. The most basic unit of observation is an ‘entry’, which is a single response variable measured at a

single time for a set of species richness levels with all other factors controlled. An ‘experiment’ is the richness

manipulation within which all other factors are controlled, but multiple response variables might have been

measured at more than one time point, thus there are often multiple entries within each experiment, and

several experiments are often published within a single study, and might differ in the level of a factor such as

consumer presence, resource supply, etc. Numbers in each column indicate the number of groups (entry,

experiment or study) in the dataset for each level of each categorical predictor, and for each continuous

predictor the range of values is given for the entire dataset.

Categorical predictors

Levels

Entry (n)

Experiments (n)

Studies (n)

Ecosystem (Sys)

Aquatic

134

Terrestrial

424

301

Trophic group (L)

Primary producers

501

327

Herbivore

Detritivore

Lab/field

Lab / greenhouse

178

121

Field enclosures or

plots

348

221

Outdoor mesocosms

Biomass estimator (Units)

Biomass

501

339

Density

Percent cover

Resource treatment (N)

Control

381

241

Addition

172

128

Reduction

Continuous predictors

Min

Median

Mean

Max

Experimental duration

0.02

1.64

8.48

202.6

Time of measurement (TG)

0.02

1.05

7.10

202.6

Smax

9.67

Accepted Article

Table 2. Hypotheses tested for how abiotic and experimental factors could affect the relationship between

species richness and biomass. Taken together, this set of hypotheses allowed us to test the overarching

hypothesis that a single BEF relationship, expressed as an empirically estimated value of b in a power

function, adequately describes the relationship between species richness and standing biomass at fine spatial

grains (e.g. m2 or litres) despite variation in abiotic conditions, sites, and ecological community contexts such

as different species pools, ecosystems, trophic levels or resource regimes.

Hypothesis

Model

There is a constant relationship between species richness and standing biomass, estimated as Β1

in Eq. 1a. Variation in unmeasured attributes at different levels of organization in the data

(entry-, experiment- or study-level variation) did not alter estimates of the richness–biomass

relationship

All, with

random

effects (Eq.

1b)

The richness–biomass relationship increases with time within experiments, estimated as

generation times of focal taxa

The effects of experimental duration within experiments varies among ecosystems and trophic

levels

The effects of experimental duration within experiments varies among trophic levels

The richness–biomass relationship varies among ecosystems (aquatic, terrestrial) and trophic

groups (primary producer, detritivore, herbivore)

In addition to variation among ecosystems and trophic groups, the richness–biomass

relationship varies with increased or reduced resources (water, nutrients, CO2)

Experiments that run for a greater number of generations show stronger effects of richness

when compared across taxa or systems, in the context of resource addition or reduction

The effects of maximum experimental duration vary with level of resource addition

The richness-biomass relationship varies with attributes of the experimental design – maximum

duration, maximum number of species tested (Smax), units in which biomass was measured

(biomass estimator), and lab vs field (model 8)

All biotic, abiotic and experimental factors (model 1–8) modify the BEF relationship (Eq. 3)

Accepted Article

When all predictors are considered, the interaction between ecosystem and trophic group is not

important (Eq. 2)

9.1

Accepted Article

Table 3. Results of model selection for basic species richness - biomass model. Models relate total estimated

biomass (ln(Y)) to species richness (ln(S)) and experimental duration, estimated in terms of number of

generations of focal taxa (ln(Tg)). Models are ranked by AICc, and compared using AIC weight (w) and AIC

values and likelihood ratio tests. Likelihood ratio tests (p-values) compare each model with the top-ranked

(lowest AICc value) model (first row) and facilitate interpretation of the significance of differences in similar

AICc values. All models include variable slope and intercept coefficients at the entry, experiment and study

level (Table A1).

Model

AICc

modLik



ln(Y) = ln(S) + ln(Tg)

1451.5

0.49

–712.67

0.00

ln(Y) = ln(S) × ln(Tg)

1452.4

0.30

–712.13

0.95

0.30

ln(Y) = ln(S)

1453.1

0.21

–714.50

1.65

0.09

Accepted Article

Table 4. Comparison of alternative models for how richness (ln(S)) affects biomass. Model terms are as shown in Equations 4a and b (intercept term not

shown here), ranked from left to right by their quality (high to low) as a description of this dataset. Models differed in fixed effects, indicated by , but all

included variable intercepts (μ0i,μ0j,μ0k) and slopes (μ1i,μ1j,μ1k) at the level of the study (i), experiment (j) and entry (k). Models were ranked based

on AICc, and differences assessed using δAIC and Akaike weights (w) and likelihood ratio tests. We used likelihood ratio test results (- p > 0.05, * p <

0.05, ** p < 0.01) to compare models with the top-ranked (lowest AICc) model only for comparisons in which one model can be derived from the other by

constraining parameter values. When this was not possible, likelihood ratio tests were not performed. A significant p-value indicates that the model with

the lower AICc value is a better description of the data. When the likelihood ratio test indicates no differences, the model with fewer parameters is

preferred.

Model

Predictor

Term

4.2

9.1

ln(S)

γ10



Time (ln(TG))

Β2



Ecosystem (Sys)

γ01









Trophic group (L)

γ02









Sys*L

γ03









Resource treatment (N)

γ07







Accepted Article

ln(max(Duration))

γ18



Lab vs field experiment

γ05





Biomass estimator (Units)

γ04





ln(Smax)

γ06





ln(S) × ln(TG)

Β3







ln(S) × Sys

γ11





ln(S) × Sys × ln(TG)

γ13









ln(S) × L

γ12





ln(S) × N

γ17







ln(S) × ln(max(Duration))

γ18



ln(S) × (lab vs field)

γ15





ln(S) × Units

γ14





ln(S) × Smax

γ16



AICc

1437.8

1442.9

1443.1

1443.5

1451.5

1452.3

1452.4

1455.4

1459.1

5.13

5.33

5.34

5.70

13.72

14.52

14.67

17.68

21.30

Accepted Article

0.784

0.060

0.054

0.045

0.001

logLik

–697.7

–696.2

–694.3

–702.4

–686.2

–712.7

–692.7

–712.1

–704.5

–710.4

Accepted Article

Table 5. Coefficients for modeled effect of richness on standing stock. Mean (+95% CI) estimate from the

best model with fixed effects (trophic level, duration, lab versus field tests and ecosystem) and variable

slopes and intercepts (model 4). Estimates give effect sizes relative to plant biomass in a terrestrial

ecosystem under nutrient control conditions. Values in bold indicate parameter estimates contributing to

the slope term that differ significantly from zero and thus modify the relationship between richness and

biomass.

Factor

Term

Model 4

Fixed

effects

Intercept

γ00

4.33 [3.45, 5.21]

ln(S)

γ10

0.26 [0.16, 0.37]

ln(TG)

Β2

0.16 [0.03, 0.28]

Ecosystem - Terrestrial

γ01

1.50 [0.48, 2.51]

L - Herbivore

γ02

–0.33 [–1.38, 0.66]

L - Detritivore

γ02

1.22 [–0.93, 3.39]

Terrestrial × Detritivore

γ03

–1.93 [–5.11, 1.26]

ln(S) × ecosystem –Terrest.

γ11

–0.07 [–0.18, 0.05]

ln(S) × L – Herbivore

γ12

0.21 [0.03, 0.38]

ln(S) × L – Detritivore

γ12

0.29 [0.01, 0.56]

ln(S) × Terrest. × Detrit.

γ13

-0.58 [-0.98, -0.17]

Random

effects

Entry – intercept

Σ0k

0.38

Entry – ln(S)

Σ1k

0.03

Experiment – intercept

Σ0j

0.72

Experiment – ln(S)

Σ1j

< 0.01

Study – intercept

Σ0i

3.31

Study – ln(S)

Σ1i

0.03

Residual

σ2

0.02

Accepted Article

Figure Legends

Figure 1. (A) Standing stock (biomass) plotted against species richness as a power function (Y = a × Sb)

relating standing biomass (Y) to species richness (S) via an intercept (a) and scaling parameter (b) for

each entry in our database (n = 558). Each entry is plotted in gray, dark lines indicate overlapping lines.

Each entry was analyzed in a hierarchical mixed effects model using a linearized power function (Eq. 1,

2). (B) Empirically estimated scaling parameters for BEF relationships vary among trophic groups and

between aquatic and terrestrial systems. Estimates are based on model coefficients for the slope term

(Β1.ijk) from the best model (model 4, Table 3; black points). Standard errors shown in this figure are

errors of the mean estimate from the distribution of fitted slopes for this dataset. Confidence intervals

estimated from the model output are shown in Table 4. (C) Power functions plotted with b values shown

in panel B for primary producers, aquatic herbivores and aquatic detritivores.

5000

10000

15000

010 20 30 40

Species Richness

Standing Biomass

Accepted Article

To the type-setter: please change x-axis legend in (a) to: Species richness

and y-axis legend to: Standing biomass

●

Aq. Primary Prod.

●

Aq. Herbivores

●

Aq. Detritovores

●

Terr. Prim. Prod

●

Terr. Detrit.

−0.2 0.0 0.2 0.4 0.6

Slope coefficients

020 40 60 80 100

0 20 40 60 80

Species richness (S)

estimated biomass (Y)

●

Plants and Algae

Aq. Herbivores

Aq. Detritivores

Terr. Detritivores

Accepted Article

Figure 2. Random effects (+ CI) estimated by the best model (model 4, Table 2) associated with entryk

(plots A–B), experimentj (panels C–D) and studyi (panels E–F) for intercepts (μ0) and slopes (μ1),

ranked by slope random effects (μ1). Gray CI’s include 0, indicating that the estimated random effect

cannot be distinguished from the fixed effect for slope or intercept. Random effects different from zero

imply that the coefficient for that study can be estimated as the fixed effect plus the random effect.

(A) Intercept random effects for Entry (μ0k) (B) Slope random effect for Entry (μ1k)

Group levels

(E) Intercept random effects for Study (μ0i) (F) Slope random effect for Study (μ1i)

Group levels

●●

●

●●

●

●●

●

●●

●

●●

●

(Intercept) logSc

−4

−2

−0.5

0.0

0.5

1.0

521130351242713081395115201307422191136510425426351519147948106881610251283134724392828130113051076180100431123072230340739530114003941198416137130438690127749532713991239102696112923911197838114547717873073912151032325129075499825583313494111183993118527912887266947078377335579054872710995581594714131401740323130067712034601743531211107349670314119571281111195654696713404087201291275751697134735018255818870884211961067171141473713421205102710709147028487173751220127943869311128010834947946118711439504784613797159361216106114969570469111068775421189459119912292561064847623314105501081850108216495367675291150947641213191325959872134411011398829834103172375578461214397135413701814391284994118612129635531184109871212241217120010634451812042571202709556710952132410801233732135311924625441228110045666912857481059540157549120167595440667210601235305839912119312429621243107130240410711372743123785736915552118113069971210729934451386700849481108474475669688852540913791062687136839610962997474171072887844735971107951483960731995140910831298114272812343528519969662745011221836840134611881109463714102993597918134567472510307182761350135110991733001179969493118212138271180673555387861241999108537869911785241000480479746817545711102812491225464122384774214081075131532448291617983010979654662654405713750120893312365511280818403123210343461403671106629811444813181065497100268512191222128912384849704655173471721299134310351110514719388348120668817523455413523023221371100112931194835698281390103639912269171767165378802912405351341128251312189581286692968119087738122769070512091412559103311951278136941341811074856898149511775088784203298323268434001831397706721701119113162788414214983041069670123140113131207749964831107424574514024301037533281037734130942328013111312415132041485251513171287741516131413105065074244295051841248148882135138504428 521130351242713081395115201307422191136510425426351519147948106881610251283134724392828130113051076180100431123072230340739530114003941198416137130438690127749532713991239102696112923911197838114547717873073912151032325129075499825583313494111183993118527912887266947078377335579054872710995581594714131401740323130067712034601743531211107349670314119571281111195654696713404087201291275751697134735018255818870884211961067171141473713421205102710709147028487173751220127943869311128010834947946118711439504784613797159361216106114969570469111068775421189459119912292561064847623314105501081850108216495367675291150947641213191325959872134411011398829834103172375578461214397135413701814391284994118612129635531184109871212241217120010634451812042571202709556710952132410801233732135311924625441228110045666912857481059540157549120167595440667210601235305839912119312429621243107130240410711372743123785736915552118113069971210729934451386700849481108474475669688852540913791062687136839610962997474171072887844735971107951483960731995140910831298114272812343528519969662745011221836840134611881109463714102993597918134567472510307182761350135110991733001179969493118212138271180673555387861241999108537869911785241000480479746817545711102812491225464122384774214081075131532448291617983010979654662654405713750120893312365511280818403123210343461403671106629811444813181065497100268512191222128912384849704655173471721299134310351110514719388348120668817523455413523023221371100112931194835698281390103639912269171767165378802912405351341128251312189581286692968119087738122769070512091412559103311951278136941341811074856898149511775088784203298323268434001831397706721701119113162788414214983041069670123140113131207749964831107424574514024301037533281037734130942328013111312415132041485251513171287741516131413105065074244295051841248148882135138504428

Group levels

●

●●

●

(Intercept) logSc

−2

−0.2

−0.1

0.0

0.1

0.2

532531154661263673024951165771663503092404634052412423834541521952844042141813024617538057446227015150456573511374805785795202495510354934343341155401450419210374434242230454120547233575165567376243162553451375366426226255386529343113610446458721646822557326760236105201274612925647613440942048542369470904121455852713603132316537837255431242758333456222754126812351922024547942352130381143652484524324815234134381741113538225356049426454052556937115017216144147731540756450040841450658433634719241381404431508467507502469453332466373448465358239459548338482455563348433489410132493379214272259144223221163499518359390355463401224564983071675505515582602635722325238523434953475262568460439266486457119576447173222149484357247218449251184583881522584783844301243648850143730155224164209160622114612515731147317150336308446490244129487483523915956144517058691389232496417595823516947154454549742249150534216813265377387492117402113614743722963412285591182242372543463705441633526926151143530334458031047225053620838311542499314174530133131257352512128217522535581153534337533 532531154661263673024951165771663503092404634052412423834541521952844042141813024617538057446227015150456573511374805785795202495510354934343341155401450419210374434242230454120547233575165567376243162553451375366426226255386529343113610446458721646822557326760236105201274612925647613440942048542369470904121455852713603132316537837255431242758333456222754126812351922024547942352130381143652484524324815234134381741113538225356049426454052556937115017216144147731540756450040841450658433634719241381404431508467507502469453332466373448465358239459548338482455563348433489410132493379214272259144223221163499518359390355463401224564983071675505515582602635722325238523434953475262568460439266486457119576447173222149484357247218449251184583881522584783844301243648850143730155224164209160622114612515731147317150336308446490244129487483523915956144517058691389232496417595823516947154454549742249150534216813265377387492117402113614743722963412285591182242372543463705441633526926151143530334458031047225053620838311542499314174530133131257352512128217522535581153534337533

Group levels

●

●●

●

●●●●

●

●●

●

●●

●

●●

●

●●

●

●●

●●●●●●

●

●●

●

(Intercept) logSc

−5.0

−2.5

0.0

2.5

5.0

−0.5

0.0

0.5

1.0

331451151311581041542115552871641613681711351471911448067161108321782414312012486251501761973721066277638522166189186597917312811428822953152651051197618414119245126160101189810716274102180231572010151371834930174843110366121168127 331451151311581041542115552871641613681711351471911448067161108321782414312012486251501761973721066277638522166189186597917312811428822953152651051197618414119245126160101189810716274102180231572010151371834930174843110366121168127

Group levels

Accepted Article

Figure 3. Expected change in biomass associated with changes in species richness. Distribution of species

richness changes (top histogram), expressed as a response ratio (ln(STime1/STime2)), and the distribution of

associated change in biomass (vertical histogram), expressed as ln(YTime1/YTime0) expected for (A) primary

producers and (B) herbivores. The distribution of expected function was produced using Y = a × Sb (the

plotted curve) for values of b = 0.26 for plants, and b = 0.47 for herbivores (Table 4, Fig. 1B). Solid blue

lines indicate response ratios of 1 = no change in richness; and the red lines indicate the mean expected

function. Dashed lines identify a 10% decline in standing biomass, and the intersection with the BEF

curve identifies the change in richness expected to cause a 10% change in function: a 35% reduction in

plant richness, and a 20% reduction in herbivore richness.

0.0

0.5

1.0

1.5

0.0 0.5 1.0 1.5 2.0

Proportion of initial richness

Proportion of initial function

0.0

0.5

1.0

1.5

0.0 0.5 1.0 1.5 2.0

Proportion of initial richness

Proportion of initial function

Accepted Article

The Equilibrium Theory of Biodiversity Dynamics: A General Framework for Scaling Species Richness and Community Abundance along Environmental Gradients

Article

Full-text available

Nov 2024

Large-scale temporal and spatial biodiversity patterns have traditionally been explained by multitudinous particular factors and a few theories. However, these theories lack sufficient generality and do not address fundamental interrelationships and coupled dynamics among resource availability, community abundance, and species richness. We propose the equilibrium theory of biodiversity dynamics (ETBD) to address these linkages. According to the theory, equilibrium levels of species richness and community abundance emerge at large spatial scales because of the population size dependence of speciation and/or extinction rates, modulated by resource availability and the species abundance distribution. In contrast to other theories, ETBD includes the effect of biodiversity on community abundance and thus addresses phenomena such as niche complementarity, facilitation, and ecosystem engineering. It reveals how alternative stable states in both diversity and community abundance emerge from these nonlinear biodiversity effects. The theory predicts how the strength of these effects alters scaling relationships among species richness, (meta)community abundance, and resource availability along different environmental gradients. Using data on global-scale variation in tree species richness, we show how the general framework is useful for clarifying the role of speciation, extinction, and resource availability in driving macroecological patterns in biodiversity and community abundance, such as the latitudinal diversity gradient.

Biodiversity loss reduces global terrestrial carbon storage

Article

Full-text available

May 2024

Natural ecosystems store large amounts of carbon globally, as organisms absorb carbon from the atmosphere to build large, long-lasting, or slow-decaying structures such as tree bark or root systems. An ecosystem’s carbon sequestration potential is tightly linked to its biological diversity. Yet when considering future projections, many carbon sequestration models fail to account for the role biodiversity plays in carbon storage. Here, we assess the consequences of plant biodiversity loss for carbon storage under multiple climate and land-use change scenarios. We link a macroecological model projecting changes in vascular plant richness under different scenarios with empirical data on relationships between biodiversity and biomass. We find that biodiversity declines from climate and land use change could lead to a global loss of between 7.44-103.14 PgC (global sustainability scenario) and 10.87-145.95 PgC (fossil-fueled development scenario). This indicates a self-reinforcing feedback loop, where higher levels of climate change lead to greater biodiversity loss, which in turn leads to greater carbon emissions and ultimately more climate change. Conversely, biodiversity conservation and restoration can help achieve climate change mitigation goals.

Functional structure of the natural enemy community of the fall armyworm, Spodoptera frugiperda in the Americas

Article

Full-text available

Oct 2024
BIOL CONTROL

An integrated approach to above- and below-ground ecological monitoring for nature-based solutions

Preprint

Full-text available

Nov 2024

1. As the development of nature-based solutions (NbS) increases globally, it is important to ensure that projects meet the objective of benefiting biodiversity, alongside tackling societal challenges. However, most NbS projects do not directly monitor ecological outcomes, and those that do often focus on a limited set of metrics. It is therefore challenging to assess whether projects fulfil the aim of benefiting biodiversity. 2. We aimed to identify the most informative and feasible ecological metrics, both above- and below-ground, for monitoring ecological outcomes of NbS. We conducted a structured non-systematic literature review to identify possible biodiversity and soil health metrics, then developed a scoring system to rank these based on their informativeness and feasibility for monitoring. 3. Metrics were categorised into compositional, structural, and functional aspects of biodiversity, and biological, physical, and chemical aspects of soil health. We grouped biodiversity and soil health metrics into Tier 1 (the most informative and feasible metrics), Tier 2 (informative metrics with some limitations in scope or feasibility), and Future metrics (highly informative metrics which are currently less feasible to monitor). Tier 1 metrics collectively address multiple aspects of biodiversity and soil health and are currently the highest priority for NbS projects to assess. For biodiversity nine Tier 1, six Tier 2, and 15 Future metrics were identified, and for soil health 11 Tier 1, six Tier 2, and five Future metrics. 4. We identified existing standardised methodologies for monitoring the proposed metrics, noting that for many metrics standardised methodologies are not available and threshold or reference values for each metric are missing. 5. Solution: Our study provides practitioners with a framework for selecting optimum metrics for assessing above- and below-ground ecological outcomes of NbS relevant to the place in which they are being implemented. A definition of each metric and standardised methodologies for collecting data are summarised, providing information to develop an ecological monitoring protocol for an NbS project. The information on each metric has been made freely available as a searchable database in an interactive online interface geared towards UK practitioners, but with wider applicability.

Ecosystem functioning during biodiversity loss and recovery

Article

Full-text available

May 2024

Anthropogenic biodiversity loss can impair ecosystem functioning. Human activities are often managed with the aim of reversing biodiversity loss and its associated functional impacts. However, it is currently unknown whether biodiversity–ecosystem function (BEF) relationships observed during biodiversity recovery are the same as those observed during biodiversity loss. This will depend on how species extirpation and recolonisation sequences compare and how different species influence ecosystem functioning. Using data from a marine benthic invertebrate community, we modelled how bioturbation potential – a proxy for benthic ecosystem functioning – changes along biodiversity loss and recovery sequences governed by species' sensitivity to physical disturbance and recolonisation capability, respectively. BEF relationships for biodiversity loss and recovery were largely the same despite species extirpation and recolonisation sequences being different. This held true irrespective of whether populations were assumed to exhibit compensatory responses as species were removed or added. These findings suggest that the functional consequences of local biodiversity loss can be reversed by alleviating its drivers, as different species present at comparable levels of species richness during biodiversity loss and recovery phases have similar functional effects. Empirically verifying and determining the generality of our model‐based results are potential next steps for future research.

Consumer biodiversity increases organic nutrient availability across aquatic and terrestrial ecosystems

Article

Oct 2024
SCIENCE

Human land-use intensification threatens arthropod (for example, insect and spider) biodiversity across aquatic and terrestrial ecosystems. Insects and spiders play critical roles in ecosystems by accumulating and synthesizing organic nutrients such as polyunsaturated fatty acids (PUFAs). However, links between biodiversity and nutrient content of insect and spider communities have yet to be quantified. We relate insect and spider richness to biomass and PUFA-mass from stream and terrestrial communities encompassing nine land uses. PUFA-mass and biomass relate positively to biodiversity across ecosystems. In terrestrial systems, human-dominated areas have lower biomass and PUFA-mass than more natural areas, even at equivalent levels of richness. Aquatic ecosystems have consistently higher PUFA-mass than terrestrial ecosystems. Our findings reinforce the importance of conserving biodiversity and highlight the distinctive benefits of aquatic biodiversity.

Precipitation differentially regulates above- and belowground productivity in response to nitrogen enrichment in an alpine meadow

Article

Oct 2024
AGR FOREST METEOROL

The fertilization effect of nitrogen (N) deposition on global productivity may be limited by water availability due to the coupling effect of N and water on ecosystem carbon cycling. However, it remains unclear whether this limitation is synchronized between above- and belowground productivity. Based on a long-term field manipulation experiment in an alpine meadow of the Qinghai-Tibetan Plateau, we found that precipitation played differential roles in regulating above- and belowground productivity responses to N addition. In experimental years with higher precipitation, N addition increased belowground productivity more. The main underlying mechanism was that higher precipitation mitigated excessive accumulation of soil inorganic N under N addition and shifted background plant community composition towards more N-favored species, which increased plant biomass allocation belowground and consequently stimulated the return of belowground productivity from N addition. In contrast, while N addition increased aboveground productivity, its effect didn’t change with precipitation. This was mainly due to the trade-off between positive and negative effects of precipitation on the N response of aboveground productivity. Specifically, the precipitation-induced changes in soil inorganic N accumulation and background plant species composition reduced the return of aboveground productivity from N addition, while precipitation also offset this effect by mitigating the N-induced plant species loss, resulting in an overall insignificant effect of precipitation on the N response of aboveground productivity. Our findings imply that the trends of belowground productivity under N deposition can’t be inferred solely from the easily measured aboveground parts, especially in the context of climate change. The work underscores an urgent need for more belowground measurements to improve the assessment of terrestrial carbon sinks.

Urgent climate action is needed to ensure effectiveness of protected areas for biodiversity benefits

Article

Sep 2024

Biodiversity promotes urban ecosystem functioning

Article

Full-text available

Jul 2024
ECOGRAPHY

The proportion of people living in urban areas is growing globally. Understanding how to manage urban biodiversity, ecosystem functions, and ecosystem services is becoming more important. Biodiversity can increase ecosystem functioning in non‐urban systems. However, few studies have reviewed the relationship between biodiversity and ecosystem functioning in urban areas, which differ in species compositions, abiotic environments, food webs, and turnover rates. We reviewed evidence of biodiversity–ecosystem functioning relationships in urban environments and assessed factors that influence the relationship direction. Based on 70 studies, relationships between biodiversity and ecosystem functioning were more positive than negative in urban areas, especially for pollination and nutrient cycling and retention. Surprisingly, positive and negative relationships between biodiversity and biomass production and storage were equally not statistically different, perhaps due to extensive plant management in urban areas. The number of studies and geographic coverage of our review was still insufficient to provide a general predictive framework for when biodiversity positively impacts ecosystem functioning. We identify gaps and opportunities to improve urban biodiversity–ecosystem functioning research and discuss how our findings can improve urban green space management.

Functional coexistence theory: a mechanistic framework linking biodiversity to ecosystem function

Preprint

Full-text available

May 2024

Theory and experiments show that diverse ecosystems often have higher levels of function (for instance, biomass production), yet it remains challenging to identify the biological mechanisms responsible. We synthesize developments in coexistence theory into a general theoretical framework linking community coexistence to ecosystem function. Our framework, which we term functional coexistence theory, identifies three components determining the total function of a community of coexisting species. The first component directly corresponds to the niche differences that enable pairwise species coexistence, and to the complementarity component from the additive partition of biodiversity effects. The second component measures whether higher functioning species also have higher competitive fitness, providing a missing link between the additive partition’s selection effect and modern coexistence theory’s concept of equalization. The third component is least well-studied: reducing functional imbalances between species increases niche difference’s positive effect on function. Using a mechanistic model of resource competition, we show that our framework can identify how traits drive the effect of competition on productivity, and confirm our theoretical expectations by fitting this model to data from a classic plant competition experiment. Furthermore, we apply our framework to simulations of communities with multiple ecosystem functions or more than two species, demonstrating that relationships between niche, fitness, and function also predict total function beyond the case studied by classical theory. Taken together, our results highlight fundamental links between species coexistence and its consequences for ecosystem function, providing an avenue towards a predictive theory of community–ecosystem feedbacks.

Biodiversity enhances ecosystem multifunctionality across trophic levels and habitats

Article

Full-text available

Apr 2015

The importance of biodiversity for the integrated functioning of ecosystems remains unclear because most evidence comes from analyses of biodiversity's effect on individual functions. Here we show that the effects of biodiversity on ecosystem function become more important as more functions are considered. We present the first systematic investigation of biodiversity's effect on ecosystem multifunctionality across multiple taxa, trophic levels and habitats using a comprehensive database of 94 manipulations of species richness. We show that species-rich communities maintained multiple functions at higher levels than depauperate ones. These effects were stronger for herbivore biodiversity than for plant biodiversity, and were remarkably consistent across aquatic and terrestrial habitats. Despite observed tradeoffs, the overall effect of biodiversity on multifunctionality grew stronger as more functions were considered. These results indicate that prior research has underestimated the importance of biodiversity for ecosystem functioning by focusing on individual functions and taxonomic groups.

Marine biodiversity and ecosystem functioning: What's known and what's next?

Article

Full-text available

Oct 2014
OIKOS

Marine ecosystems are experiencing rapid and pervasive changes in biodiversity and species composition. Understanding the ecosystem consequences of these changes is critical to effectively managing these systems. Over the last several years, numerous experimental manipulations of species richness have been performed, yet existing quantitative syntheses have focused on a just a subset of processes measured in experiments and, as such, have not summarized the full data available from marine systems. Here, we present the results of a meta‐analysis of 110 marine experiments from 42 studies that manipulated the species richness of organisms across a range of taxa and trophic levels and analysed the consequences for various ecosystem processes (categorised as production, consumption or biogeochemical fluxes). Our results show that, generally, mixtures of species tend to enhance levels of ecosystem function relative to the average component species in monoculture, but have no effect or a negative effect on functioning relative to the ‘highest‐ performing’ species. These results are largely consistent with those from other syntheses, and extend conclusions to ecological functions that are commonly measured in the marine realm (e.g. nutrient release from sediment bioturbation). For experiments that manipulated three or more levels of richness, we attempted to discern the functional form of the biodiversity–ecosystem functioning relationship. We found that, for response variables related to consumption, a power‐function best described the relationship, which is also consistent with previous findings. However, we identified a linear relationship between richness and production. Combined, our results suggest that changes in the number of species will, on average, tend to alter the functioning of marine ecosystems. We outline several research frontiers that will allow us to more fully understand how, why, and when diversity may drive the functioning of marine ecosystems. Synthesis The oceans host an incredible number and variety of species. However, human activities are driving rapid changes in the marine environment. It is imperative we understand ecosystem consequences of any associated loss of species. We summarized data from 110 experiments that manipulated species diversity and evaluated resulting changes to a range of ecosystem responses. We show that losing species, on average, decreases productivity, growth, and a myriad of other processes related to how marine organisms capture and utilize resources. Finally, we suggest that the loss of species may have stronger consequences for some processes than others.

Global meta-analysis reveals no net change in local-scale plant biodiversity over time

Article

Full-text available

Oct 2013
P NATL ACAD SCI USA

Significance A major advance of the last 20 y at the interface of biological, environmental, and conservation sciences has been the demonstration that plant biodiversity positively influences ecosystem function. Linking these results to applied conservation efforts hinges on the assumption that biodiversity is actually declining at the local scale at which diversity–function relationships are strongest. Our compilation and analysis of a global database of >16,000 repeat survey vegetation plots from habitats across the globe directly contradict this assumption. We find no general tendency for local-scale plant species diversity to decline over the last century, calling into question the widespread use of ecosystem function experiments to argue for the importance of biodiversity conservation in nature.

Effects of predator richness on prey suppression: A meta-analysis

Article

Full-text available

Oct 2013

It is well established that species richness of primary producers and primary consumers can enhance efficiency of resource uptake and biomass production of respective trophic levels. At the level of secondary consumers (predators), however, conclusions about the functional role of biodiversity have been mixed. We take advantage of a recent surge of published experiments (totaling 46 since 2005) to both evaluate general effects of predator richness on aggregate prey suppression (top-down control) and explore sources of variability among experiments. Our results show that, across experiments, predator richness enhances prey suppression relative to the average single predator species (mean richness effect), but not the best-performing species. Mean richness effects in predator experiments were stronger than those for primary producers and detritivores, suggesting that relationships between richness and function may increase with trophic height in food webs. The strength of mean predator richness effects increased with the spatial and temporal scale of experiments, and the taxonomic distinctness (TD, used as a proxy of phylogenetic diversity) of species present. This latter result suggests that TD captures important aspects of functional differentiation among predators and that measures of biodiversity that go beyond species richness may help to better predict the effects of predator species loss.

The relationship between body mass and field metabolic rate among individual birds and mammals

Article

Full-text available

May 2013
J ANIM ECOL

The power-law dependence of metabolic rate on body mass has major implications at every level of ecological organization. However, the overwhelming majority of studies examining this relationship have used basal or resting metabolic rates, and/or have used data consisting of species-averaged masses and metabolic rates. Field metabolic rates are more ecologically relevant and are probably more directly subject to natural selection than basal rates. Individual rates might be more important than species-average rates in determining the outcome of ecological interactions, and hence selection. We here provide the first comprehensive database of published field metabolic rates and body masses of individual birds and mammals, containing measurements of 1498 animals of 133 species in 28 orders. We used linear mixed-effects models to answer questions about the body mass scaling of metabolic rate and its taxonomic universality/heterogeneity that have become classic areas of controversy. Our statistical approach allows mean scaling exponents and taxonomic heterogeneity in scaling to be analysed in a unified way while simultaneously accounting for nonindependence in the data due to shared evolutionary history of related species. The mean power-law scaling exponents of metabolic rate vs. body mass relationships were 0·71 [95% confidence intervals (CI) 0·625–0·795] for birds and 0·64 (95% CI 0·564–0·716) for mammals. However, these central tendencies obscured meaningful taxonomic heterogeneity in scaling exponents. The primary taxonomic level at which heterogeneity occurred was the order level. Substantial heterogeneity also occurred at the species level, a fact that cannot be revealed by species-averaged data sets used in prior work. Variability in scaling exponents at both order and species levels was comparable to or exceeded the differences 3/4−2/3 = 1/12 and 0·71−0·64. Results are interpreted in the light of a variety of existing theories. In particular, results are consistent with the heat dissipation theory of Speakman & Król (2010) and provided some support for the metabolic levels boundary hypothesis of Glazier (2010). Our analysis provides the first comprehensive empirical analysis of the scaling relationship between field metabolic rate and body mass in individual birds and mammals. Our data set is a valuable contribution to those interested in theories of the allometry of metabolic rates. The authors provide the first comprehensive empirical analysis of the scaling relationship between field metabolic rate and body mass in individual birds and mammals. The analysis reveals the importance of heterogeneity in the scaling exponent, with consequences for biomass and nutrient flow through communities, and the structure and functioning of whole ecosystems.

Estimating local biodiversity change: A critique of papers claiming no net loss of local diversity

Article

Mar 2016
ECOLOGY

Global species extinction rates are orders of magnitude above the background rate documented in the fossil record. However, recent data syntheses have found mixed evidence for patterns of net species loss at local spatial scales. For example, two recent data meta-analyses have found that species richness is decreasing in some locations and is increasing in others. When these trends are combined, these papers argued there has been no net change in species richness, and suggested this pattern is globally representative of biodiversity change at local scales. Here we reanalyze results of these data syntheses and outline why this conclusion is unfounded. First, we show the datasets collated for these syntheses are spatially biased and not representative of the spatial distribution of species richness or the distribution of many primary drivers of biodiversity change. This casts doubt that their results are representative of global patterns. Second, we argue that detecting the trend in local species richness is very difficult with short time series and can lead to biased estimates of change. Reanalyses of the data detected a signal of study duration on biodiversity change, indicating net biodiversity loss is most apparent in studies of longer duration. Third, estimates of species richness change can be biased if species gains during post-disturbance recovery are included without also including species losses that occurred during the disturbance. Net species gains or losses should be assessed with respect to common baselines or reference communities. Ultimately, we need a globally coordinated effort to monitor biodiversity so that we can estimate and attribute human impacts as causes of biodiversity change. A combination of technologies will be needed to produce regularly updated global datasets of local biodiversity change to guide future policy. At this time the conclusion that there is no net change in local species richness is not the consensus state of knowledge.

Model Selection and Multimodel Inference

Article

Jan 2004

This chapter gives results from some illustrative exploration of the performance of information-theoretic criteria for model selection and methods to quantify precision when there is model selection uncertainty. The methods given in Chapter 4 are illustrated and additional insights are provided based on simulation and real data. Section 5.2 utilizes a chain binomial survival model for some Monte Carlo evaluation of unconditional sampling variance estimation, confidence intervals, and model averaging. For this simulation the generating process is known and can be of relatively high dimension. The generating model and the models used for data analysis in this chain binomial simulation are easy to understand and have no nuisance parameters. We give some comparisons of AIC versus BIC selection and use achieved confidence interval coverage as an integrating metric to judge the success of various approaches to inference.

The Ecological Implications of Body Size

Article

Apr 1985

The biodiversity-dependent ecosystem service debt

Article

Nov 2014
ECOL LETT

Habitat destruction is driving biodiversity loss in remaining ecosystems, and ecosystem functioning and services often directly depend on biodiversity. Thus, biodiversity loss is likely creating an ecosystem service debt: a gradual loss of biodiversity-dependent benefits that people obtain from remaining fragments of natural ecosystems. Here, we develop an approach for quantifying ecosystem service debts, and illustrate its use to estimate how one anthropogenic driver, habitat destruction, could indirectly diminish one ecosystem service, carbon storage, by creating an extinction debt. We estimate that c. 2-21 Pg C could be gradually emitted globally in remaining ecosystem fragments because of plant species loss caused by nearby habitat destruction. The wide range for this estimate reflects substantial uncertainties in how many plant species will be lost, how much species loss will impact ecosystem functioning and whether plant species loss will decrease soil carbon. Our exploratory analysis suggests that biodiversity-dependent ecosystem service debts can be globally substantial, even when locally small, if they occur diffusely across vast areas of remaining ecosystems. There is substantial value in conserving not only the quantity (area), but also the quality (biodiversity) of natural ecosystems for the sustainable provision of ecosystem services. © 2014 John Wiley & Sons Ltd/CNRS.

Coexistence, niches and biodiversity effects on ecosystem functioning

Article

Dec 2012
ECOL LETT

General principles from coexistence theory are often invoked to explain how and why mixtures of species outperform monocultures. However, the complementarity and selection effects commonly measured in biodiversity experiments do not precisely quantify the niche and relative fitness differences that govern species coexistence. Given this lack of direct correspondence, how can we know whether species-rich mixtures are stable and that the benefits of diversity will therefore persist? We develop a resource-based included-niche model in which plant species have asymmetric access to a nested set of belowground resource pools. We use the model to show that positive complementarity effects arise from stabilising niche differences, but do not necessarily lead to stable coexistence and hence can be transient. In addition, these transient complementarity effects occur in the model when there is no complementary resource use among species. Including a trade-off between uptake rates and the size of the resource pool stabilised interactions and led to persistent complementarity coupled with weak or negative selection effects, consistent with results from the longest-running field biodiversity experiments. We suggest that future progress requires a greater mechanistic understanding of the links between ecosystem functions and their underlying biological processes.