SCIENCE ADVANCES | RESEARCH ARTICLE

ECOLOGY Copyright © 2019

The Authors, some
Tropical fish diversity enhances coral reef functioning rights reserved;
exclusive licensee
across multiple scales American Association
for the Advancement
of Science. No claim to
Jonathan S. Lefcheck1,2, Anne A. Innes-Gold3, Simon J. Brandl4, Robert S. Steneck5, original U.S. Government
Ruben E. Torres6, Douglas B. Rasher1* Works. Distributed
under a Creative
There is now a general consensus that biodiversity positively affects ecosystem functioning. This consensus, how- Commons Attribution
ever, stems largely from small-scale experiments, raising the question of whether diversity effects operate at mul- NonCommercial
tiple spatial scales and flow on to affect ecosystem structure in nature. Here, we quantified rates of fish herbivory License 4.0 (CC BY-NC).
on algal turf communities across multiple coral reefs spanning >1000 km of coastline in the Dominican Republic.
We show that mass-standardized herbivory rates are best predicted by herbivore biomass and herbivore species
richness both within (-diversity) and across sites in the region (-diversity). Using species-diversity models, we
demonstrate that many common grazer species are necessary to maximize the process of herbivory. Last, we link
higher herbivory rates to reduced algal turf height and enhanced juvenile coral recruitment throughout the eco-
system. Our results suggest that, in addition to high herbivore biomass, conserving biodiversity at multiple scales
is important for sustaining coral reef function.

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

INTRODUCTION removal of herbivorous fishes from coral reefs has led to state shifts
The idea of biological diversity as both a response to and a driver of in many regions, particularly in the Caribbean, where decades of
ecosystem processes has led to parallel tracks of investigation over overfishing, temperature-induced coral bleaching, and disease events
the past century. On the one side, macroecological research has have fostered algal dominance (15). The role of biodiversity loss in
focused on the origin and maintenance of diversity, and how com- mediating this transition, however, remains unclear, leading to the
munities of species come together across space and time. Integral to suggestion that high biomass of only a few select key species are
this approach is the partitioning of diversity into local  and regional needed to sustain or rebuild reef function (16). At the same time,
 components, with a third component——that quantifies compo- several recent studies have shown that different herbivores target
sitional variation among local communities within a region (1). A different algal resources (17–19), creating the potential for strong
second field relating biodiversity to the processes that underpin complementarity in these ecosystems. This new experimental evi-
functioning ecosystems has now unequivocally demonstrated that dence raises questions regarding whether and how diversity effects
the loss of species leads to measurable declines in many processes manifest, whether they are broadly generalizable, whether they rival
such as biomass accumulation, nutrient cycling, and decomposition the importance of high herbivore biomass, and, critically, whether
(2, 3). A major criticism of these studies, however, is their limited their contributions to herbivory generate measurable consequences
scope: As most studies are limited to analysis of a single location, they for the reef itself (17). Moreover, while some herbivory studies
often ignore the larger regional pool from which patterns in local di- have revealed a dominant effect of herbivore identity at discrete
versity arise (4). Recent theoretical (5, 6) and experimental (7–11) locations (20), such identity effects might combine and give rise to
studies have linked local- and landscape-scale effects of biodiversity an emergent effect of diversity at larger scales of space and time, as
using metacommunity dynamics, and new efforts to “scale-up” bio- has been shown in terrestrial ecosystems (21, 22). If so, many more
diversity research in the real world have leveraged observational species might be needed to maintain ecosystem functioning across
datasets across large spatial and environmental gradients (3). Still, large, naturally varied reefscapes than would be expected from
few analy­ses have incorporated an explicit regional context to explain existing evidence.
local functioning (12–14), although doing so would help to recon- To evaluate the relationship between herbivore diversity and the
cile the demonstrated benefits of biodiversity with the scales at process of herbivory, we deployed remote video cameras on 10 reefs
which natural resources are often managed. spanning >1000 km of coastline in the Dominican Republic and
Here, we evaluate biodiversity-ecosystem functioning relation- quantified herbivore grazing rates on the benthic algal turf commu-
ships at various scales by relating patterns of tropical fish - and nity. We then predicted mass-standardized bite rates [i.e., bites
-diversity to herbivory rates on Dominican coral reefs. Herbivory multiplied by the biomass of the herbivore, given that per capita
is a key ecological process on reefs globally, where intense grazing feeding impacts scale with fish body size (20)] as a function of total
by herbivores prevents the establishment and accumulation of algae herbivore biomass, local -diversity and within-region -diversity,
that can suppress coral growth, survival, and reproduction (15). The co-occurring bottom-up drivers such as resource availability, and
unmeasured site-to-site differences (via random effects). Here, we use
1
Bigelow Laboratory for Ocean Sciences, East Boothbay, ME 04544, USA. 2Tennenbaum
the compositional uniqueness of each community relative to the re-
Marine Observatories Network, MarineGEO, Smithsonian Environmental Research gional species pool as a measure of local contributions to -diversity
Center, Edgewater, MD 21037, USA. 3Vassar College, Poughkeepsie, NY 12604, USA. (LCBD) (23). This approach provides a unique value of -diversity
4
Simon Fraser University, Burnaby, BC V5A 1S6, Canada. 5Darling Marine Center, for each site—increasing our statistical power—with larger values
University of Maine, Walpole, ME 04573, USA. 6Reef Check Dominican Republic,
Santo Domingo, DN 10148, Dominican Republic. indicating greater compositional differences relative to other sites.
*Corresponding author. Email: drasher@bigelow.org Last, to assess the cascading consequences of changing herbivore

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 1 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

biomass and richness at the ecosystem level, we evaluated how ob- does not modify the contributions of any one species on average
served herbivory rates from our video assays predicted measures of (Table 1). However, this interpretation should be met with caution,
benthic community structure on each reef (derived from indepen- as these models consider only two-way interactions and linearized
dent transect surveys). relationships (25), whereas actual interactions in nature may mani-
fest between many species and may occur nonlinearly (28). More-
over, species-diversity models are also sensitive to sample size, and
RESULTS AND DISCUSSION while some species appear to contribute substantially to grazing rate
We found that herbivorous fish community biomass, -diversity, when they are present (e.g., Scarus taeniopterus, Scarus vetula, and
and -diversity (LCBD) all significantly predicted mass-standardized Sparisoma viride; Fig. 2), they may not have been sufficiently abun-
grazing rate using a general linear mixed-effects model (Fig. 1 and dant across the study sites (e.g., because of high fishing pressure at
table S1). The model, which also included the (nonsignificant) in- some sites) to generate a significant effect in our model (Table 1).
fluence of coral abundance, turf algae abundance, and sea urchin Thus, while potentially conservative, these model results suggest
(Diadema antillarum) abundance (24) on rates of herbivory, ex- that many species underpin the process of herbivory at the local
plained R2 = 80% of the variance in grazing rate when considering scale, thereby creating a positive effect of herbivore -diversity on
fixed effects only, and 92% when considering both fixed and ran- ecosystem function.
dom effects. Comparison of standardized effect sizes revealed that Our finding that herbivory is positively associated with -diversity
herbivore biomass was the strongest predictor of herbivory rate (Fig. 1C) has important implications for understanding diversity
(std = 0.74). However, - and -diversity also had strong, indepen- effects at scales beyond local observations. Such a finding is consist­

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

dent impacts beyond that of biomass (std = 0.48 and 0.53 for  and , ent with the “spatial insurance hypothesis” (5), in which regional
respectively). Partial effects plots, which isolate the independent ef- biodiversity becomes important across a varied landscape because
fect of each predictor after accounting for the influence of all other the dominant contributors differ from site to site (21). This mecha-
predictors in the multiple regression model, demonstrate that, for nism has recently been revealed at similar spatial scales in analyses
a given level of herbivore biomass and resource availability, more of insect pollinator (29) and grassland diversity (22). Our findings
diverse and more compositionally unique herbivore assemblages also contextualize previous studies of herbivory on coral reefs,
are each associated with more intense grazing on the reef (Fig. 1, which often find different herbivores to be the dominant agents
B and C). of herbivory on individual reefs [e.g., (20)], by identifying a region-
The biomass of one or a few functionally important species is al biodiversity effect (i.e., “spatial dominance”) as the underlying
often considered the primary determinant of herbivory (20). To driver of that pattern. Our findings therefore indicate that, in addi-
elucidate which herbivores significantly contributed to herbivory, tion to high herbivore richness at the local scale (Fig. 1B), a diverse
we used species-diversity models to quantify the unique contribu- assemblage at the regional level may also be required to sustain
tion of each species to, as well as their average pairwise interactive ecosystem functioning across the heterogeneous reefscape, in part
effects on, the total grazing rate (25). Of the nine species observed in because different species come to dominate herbivory at different
our study, four were found to independently enhance this process sites.
(Table 1). Previous studies have interpreted such a result as evidence To evaluate whether the top-down biodiversity effects we ob-
for species complementarity (25), or the idea that each species con- served in nature are associated with enhanced ecosystem structure—a
tributes differently and additively to a process. Feeding comple- question rarely explored in biodiversity-function research—we sta-
mentarity is the most parsimonious explanation for why grazing tistically compared each estimate of herbivory to the benthic
rates scale positively with local herbivore richness (Fig. 1B), as composition of the surrounding reef (derived from independent
recent field studies have shown that herbivorous fish species finely reef-scale transect surveys). Here, we focused on the components of
partition the niche by each targeting different components of the the benthos most closely linked to the process of grazing, namely,
algal turf community and/or habitat features when feeding (26, 27). the degree to which algal turf communities were cropped, and the
The lack of a significant interaction term suggests that these effects knock-on effects of algal turf canopy height to juvenile coral densi-
do not arise synergistically—that is, the presence of another species ty (30). We found that, after accounting for the influence of coral

50
A B 30 C
kg bites m–2 h–1 | Z

25 20 20

0 10
0 0
−25
−10
−50
−20
−20
−2 0 2 −1.0 −0.5 0.0 0.5 1.0 −0.01 0.00 0.01 0.02
Biomass | Z α-Diversity | Z β-Diversity | Z

Fig. 1. Herbivore biomass, local -diversity, and between-community -diversity significantly predict mass-standardized herbivory rates. Plotted values are the
partial effects, which, having accounted for the influence of all other predictors (Z) in the linear mixed-effects model, thus reflect the statistically independent effect of
herbivore (A) biomass, (B) -diversity, and (C) -diversity on the response (mass-standardized bite rate). Fitted lines are linear regressions ± 95% confidence intervals.
Points in (C) are scaled by local herbivore richness so that larger points reflect sites with more species. The full model results are found in table S1.

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 2 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

Table 1. Results from a species-diversity model regressing mass-standardized bite rate against the proportional biomass of each species at each site,
as well as their average pairwise interaction. The average pairwise interaction was obtained by computing the product of the relative biomass of each
species and then summing these products. The model explained R2 = 90% of the variation in local herbivory rate.
Species Estimate SE t P
Acanthurus bahianus 2.257 0.576 3.918 0.001
Acanthurus coeruleus 3.070 1.260 2.437 0.027
Scarus iseri 1.957 0.715 2.735 0.015
Scarus taeniopterus 2.608 3.372 0.773 0.451
Scarus vetula 0.970 0.883 1.098 0.288
Sparisoma aurofrenatum 1.282 0.375 3.419 0.004
Sparisoma chrysopterum 0.783 2.056 0.381 0.708
Sparisoma rubripinne −0.792 3.072 −0.258 0.800
Sparisoma viride 1.164 0.828 1.406 0.179
Average interaction 0.662 0.421 1.572 0.136

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

Richness Our finding that local and regional biodiversity appears to affect
2 4 5 6 6 6 6 7 7 7
150 ecosystem functioning is timely, as several global syntheses have re-
vealed a net loss of local species richness in tropical ecosystems over
recent timescales (34, 35). While all of the species in our study occur
Cumulative bites h–1

A. bahianus
throughout the Caribbean basin—thus forming a common regional
100 A. coeruleus pool of species from which local diversity can arise—stochastic pro-
S. iseri cesses and human activities have altered species richness on both a
S. taeniopterus regional and reef-to-reef basis, with many harvested species [e.g.,
the large-bodied parrotfishes, which exert particularly strong im-
S. vetula
50
Sp. aurofrenatum
Sp. chrysopterum pacts (31)] now being rare or absent from heavily fished reefs
Sp. rubripinne (Fig. 2). Such differences in local-scale richness were easy to detect
using remote video assays and were a strong predictor of local
Sp. viride

0 herbivory rate (in addition to fish biomass). One synthesis further

noted that in locations where -diversity has not changed in recent
time, this stasis was often countered by a significant shift in species
To hia

Pe a
or lc n
or ar o

1
Ba to 2
nc o B ion
ua e

os o
um
El tug
C Pa no
C l G nit

C tus
ol rad
es n
G n

Ba nc rat
Ba

R rde
al de

se
ia

u
r

composition, i.e., -diversity. Recent compositional changes in the

a
rer
To

Caribbean are striking: The largest parrotfishes (Scarus guacamaia,

o
a

Fig. 2. Contributions to total bite rate by each grazer species at each reef. Val- Scarus coelestinus, and Scarus coeruleus) were extirpated from most
ues are averaged across all cameras at the 10 reef sites (primary x axis). Reef sites locales over the past century (and were thus absent from our study),
are in order of increasing herbivore species richness (secondary x axis). and other large-bodied species (e.g., Sparisoma viride and Scarus
vetula) are now rare on heavily fished reefs (36). Ultimately, while
cover and sea urchin (D. antillarum) abundance, the mean grazing our study provides new insight into the ecological consequences of
rate was negatively correlated with the canopy height of the algal reduced consumer richness and altered community composition
turf community (P = 0.03) (Fig. 3A). In turn, algal turf canopy on Caribbean reefs, it was conducted on a “shifted baseline” and is
height was negatively associated with the density of juvenile corals therefore likely to underestimate the true impact of historical bio-
on the reef (P = 0.03) (Fig. 3B), a link (i.e., competitive interaction) diversity loss in the region.
that has been causally demonstrated elsewhere in the Caribbean Note that, although our study was replicated over >1000 km of
(30, 31). A doubling of turf height from 2 to 4 mm predicted 10% coastline, our observations of herbivory still occurred in 1-m2 plots.
lower coral recruitment, whereas a quadrupling to 8-mm canopy However, emerging evidence indicates that an area this size is one
height was predicted to result in 30% fewer recruits (Fig. 3B). Thus, of the scales at which turf-cropping herbivores partition the niche
our findings imply a diversity-mediated cascade, wherein diverse on coral reefs. The algal turf community considered here, while
herbivore assemblages more effectively crop the reef, in turn creating superficially homogenous, is actually a consortium of filamentous
a more hospitable environment for coral settlement and survival, algae, crustose coralline algae, seaweed germlings, microorganisms,
ultimately enhancing reef integrity. Although the components of this detritus, and a variety of endolithic resources. As a result, each 1-m2
cascade are well established (15) and were recently corroborated at area of this community can represent a diverse suite of resources
similar spatial scales (32), previous studies have generally focused with differing nutritional and defensive properties. Different grazer
only on fish biomass and did not consider the instigating role of species consume these resources in a complementary fashion,
biodiversity in this process. The importance of herbivore diversity through targeting different taxa (37) and spatially partitioning their
in this cascade has clear implications for fisheries management and feeding across microtopographic features [the millimeter to cen-
reef conservation (32, 33). timeter scale (27)] and among vertical versus horizontal surfaces

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 3 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

0.4
A 40
B ditional approaches: They are less intrusive than diver surveys; allow

2
one to directly quantify the ecological process of interest (herbivory

Juvenile corals m
Turf height | Z

0.2 30
rate) rather than infer it from community attributes (e.g., standing
0.0
20 stock proxies such as algal or herbivore abundance); and allow more
10
accurate estimates of herbivory compared to diver follows, because
−0.2 each foraging bout can be slowed down or reviewed during playback
0 to ensure correct scoring. We studied all reefs within a 2-week period
−0.4 to limit confounding variation in abiotic factors (e.g., season) that
−50 −25 0 25 50
2 4 6 8 may affect rates of herbivory or benthic condition. Last, we used a
kg bites m 2 h 1 | Z Turf height (mm)
general linear mixed-effects model to quantify the independent
Fig. 3. Higher herbivore bite rates were associated with more finely cropped effects of herbivore community biomass, herbivore diversity, and
turfs, which would otherwise reduce the recruitment of corals to the reef. habitat characteristics on herbivory.
Plotted values in (A) are the partial effects of bite rate, which, having accounted for
the influence of all other predictors (Z) in the multiple regression model, thus re-
flect the statistically independent effect of mass-standardized bite rate (kg bites per
Quantifying benthic community structure
m2 per hour) on algal turf canopy height (mm) (R2 = 0.80). (B) Bivariate correlation We used a modified Atlantic and Gulf Rapid Reef Assessment
between algal turf canopy height and the number of juvenile corals per m2 (R2 = (AGRRA) protocol to quantify sessile benthic community structure
0.48). Fitted lines are linear regressions ± 95% confidence intervals. on each reef. These surveys were performed to characterize the reef
as well as to gauge whether variation in herbivore feeding is associ-

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

ated with notable differences in reef condition. At each site, four
[the centimeter to meter scale (26)]. These, too, are the small scales replicate 10-m transect lines were deployed randomly on the reef
at which algal-coral competition and coral settlement occur on (minimum of 10-m spacing between each). The number of centi-
the reef (30). Thus, small plots are one of several scales at which to meters on the tape intercepted by live coral (measured for each
investigate herbivore effects in this ecosystem, and may be conser- species), sponges, gorgonians, and benthic algae [measured by
vative when considering the additional axes of resource partitioning functional group: filamentous turf algae, encrusting coralline algae
that are known to arise among herbivores at larger scales (26). (Corallinales), non-coralline (peyssonnelid) chip-like crusts, articu-
The Caribbean appears more susceptible to state shifts from cor- lated coralline algae, and upright fleshy macroalgae] was recorded.
al to algal dominance than do other tropical regions (38). It is also For each species/functional group found on a given transect, we
among the most species-poor regions of the tropics in terms of her- counted the number of centimeters in which it was intercepted and
bivorous fish richness, raising the question of whether Caribbean divided that number by the total transect length to calculate percent
reefs are particularly sensitive to species loss. Our analysis indicates cover. On each transect, we also measured the canopy height (to the
that most common grazer species in the Caribbean are critical to the nearest millimeter) of the algal turf community using a ruler (n = 5
process of herbivory. This result reinforces the notion that herbivore to 10 measurements per transect); these measurements were then
communities exhibit little functional redundancy if examined at averaged to produce a transect-level estimate (n = 4 per reef). Coral
sufficient resolution (19, 26, 27, 37). Therefore, measures that foster species were aggregated into total coral cover for analysis. Filamen-
both herbivore biomass and species diversity at local and regional tous turf algae and crustose coralline algae were combined as the
spatial scales are likely to be more effective in rebuilding Caribbean total algal turf cover in our analysis, since both (as well as the detritus,
reef resilience than will approaches focused solely on herbivore seaweed germlings, and cyanobacteria that reside in the turf) are
abundance, biomass, or identity. part of the “epilithic algal matrix” targeted by scraping and crop-
ping herbivores. All upright fleshy macroalgae, articulated coralline
algae, and peyssonnelids were aggregated as total macroalgal cover
MATERIALS AND METHODS in our analysis.
Experimental design We quantified the density of juvenile corals on the reef by de-
In May 2017, we studied 10 reefs among six locations that span ploying a 25 cm × 25 cm quadrat at five intervals (0-, 2.5-, 5-, 7.5-,
>1000 km of coastline in the Dominican Republic (fig. S1). We and 10-m marks) along each benthic transect. At each interval, the
selected these reefs because they vary in herbivore richness and bio- quadrat was placed on the nearest hard reef substratum largely de-
mass (likely owing to variation in fishing pressure), but otherwise void of adult coral (i.e., <25% cover of live coral). Operationally, we
were of the same depth and reef type and are composed of species defined juvenile corals as those 10 to 40 mm in diameter; individuals
from the same regional pool of flora and fauna. Reef-to-reef differ- of this size have already “run the recruitment gauntlet” and there-
ences in herbivore community structure were randomly distributed fore may, with time, contribute to the adult population (30). Juvenile
across the study range (i.e., they did not follow some underlying corals 10 to 40 mm in size are not, however, large enough to influ-
oceanographic gradient). Moreover, any potential site-to-site dif- ence herbivore biomass or richness. Each juvenile coral found in the
ferences in benthic community structure or environmental condi- quadrat was identified to species and measured to the nearest milli-
tion, while small, were accounted for in our model (see below). At meter. Site-level means were generated from the 20 quadrats per-
each location (except La Caleta and Pedernales, where we only studied formed on the reef.
one reef), reefs were separated by ~1.5 to 7 km. Locations were sepa- Last, the density and sizes of the sea urchin D. antillarum were
rated by ~50 to 250 km (fig. S1). At each reef, we deployed video quantified within two belt transects (each 1 m × 10 m) on either side
cameras to capture the process of herbivory and conducted SCUBA-­ of the transect tape. Thus, within each transect, we surveyed a
based visual transect surveys to characterize benthic community 20-m2 area (n = 4 per reef). We note that this urchin was function-
structure. Camera assays provide a number of benefits beyond tra- ally absent from all but two sites in our study.

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 4 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

Quantifying herbivory rate (thereby not contributing to the process of interest). Four seconds
We used video cameras to quantify rates of herbivore grazing on the reflected a natural breakpoint in the data for fishes that did not take
algal turf community at each site. We deployed cameras (Hero 3 a bite, and was well below the average length of time that individuals
and Hero 4, GoPro Inc.) at three haphazardly selected locations at who did feed spent in the plot (21.9 s). These procedures left a total
the same depth (8 to 10 m) around each reef. Cameras were distrib- of n = 759 observations (feeding bouts) in our final dataset.
uted at least 30 to 50 m apart (i.e., comparable in total spread to that
of a traditional fish transect survey); this distribution likely exceeded Data preparation
the home range size of some, but not all, herbivorous fish species. We aggregated the benthic survey data at the reef level to produce
However, given that we were specifically interested in quantifying an average site-level abundance (% cover) for corals, macroalgae,
the aggregate effects of grazing observed in each plot through time, and the algal turf community and to produce an average estimate of
any individuals with large home ranges that visited multiple plots algal turf canopy height, urchin density, and juvenile coral density
(or a single plot multiple times) do not confound our results, but for each site. For each fish observed in our video assay, we estimated
rather intentionally reflect the natural, cumulative impacts of the its biomass using an established, species-specific length to weight
herbivore community in each plot of reef over time. We positioned relationship. Total herbivore abundance (i.e., density) and biomass
each camera so that its field of view was focused on a flat, 1-m2 area observed in each assay were computed by summing the individual
of hard (calcium carbonate) continuous reef substrate. In each case, counts and biomasses of each species observed in the assay plot,
the algal turf community occupied at least 75% of the plot. Patches respectively. To compute the mass-standardized bite rate [i.e., a
of sand or rubble, while rare, were avoided. At the beginning of the measure of herbivory that incorporates the known positive influ-

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

video, the diver indicated the 1-m2 area in the frame with a meter ence of herbivore size on feeding impact (20)], we multiplied every
stick or tape, after which point the plots were left alone for an aver- bite taken by an individual fish by its biomass (kg × bites) before
age of 43 ± 7 min (mean ± SD), the timing of which varied depend- totaling all bites for each species and dividing by the duration of the
ing on daily logistical constraints. Several replicates were discarded video footage. Bite rates were then multiplied by 60 min to yield
because of poor video quality, resulting in n = 26 cameras capturing mass-standardized bites per m2 per hour.
1049 min or ~17.5 hours of observations across all reef sites. Local -diversity was calculated as the total number of unique
To ensure that our videos were of sufficient duration to rigor- species that grazed in each video assay. We chose to compute diver-
ously estimate rates of herbivory at each site (and to investigate the sity from the camera footage rather than using SCUBA-based visual
potential biases associated with using video footage of varying surveys because these values (as well as associated abundance and
length), we constructed accumulation curves for both the number biomass estimates) reflect the diversity of fishes that actually inter-
of bites and the total species richness observed within each video. acted with each plot and contributed to the process of herbivory. To
We then applied a change point analysis to identify the exact time calculate LCBD, we first constructed a community distance matrix
point at which the curves saturated (see supplementary code). Both based on presence-absence data from each assay using the Jaccard
the number of bites and the number of species present in a video dissimilarity index. We then partitioned the dissimilarity matrix
saturated after an average of 18 and 17 min, respectively (figs. S2 into the “richness difference” component, which quantifies the dif-
and S3), indicating that our assays (mean, 43 min) were of more ferences in community composition that arise from depauperate
than sufficient duration to capture the process of herbivory. sites being nested subsets of more speciose sites, and the “replace-
When scoring each video on the computer, the 1-m2 plot in the ment” component, which identifies species or sets of species that
field of view was traced onto transparency paper placed over the turn over along a predefined gradient (39). We analyzed both com-
computer screen and marked in 10-cm increments. Scoring of ponents in our model, but only the richness difference component
herbivory began 30 s after all human activity ceased in the field of was significant (table S1), confirming our initial supposition that we
view and fishes resumed their normal swimming and feeding be- had not inadvertently captured some underlying gradient (i.e., less
havior, a period usually lasting 1 to 5 min. No humans were diving speciose sites were nested subsets of the most diverse sites rather
in the vicinity of the assay, other than for deployment and retrieval. than wholly different sets of species). We then computed indices of
Each time a nominally herbivorous fish—here, parrotfishes (family LCBD as our index of -diversity using the square root of the dis-
Labridae) and surgeonfishes (family Acanthuridae)—entered the similarity matrices for the richness and replacement components
plot and first took a bite, we identified the individual to species and (23, 39). The benefits of the LCBD approach are that it (i) provides
estimated its length to the nearest 5-cm size increment. We then a unique measure of -diversity for each site, increasing statistical
enumerated the total number of bites each individual took while in power, and (ii) is not derived from, and therefore is statistically
the frame. A bite was counted as any time a fish struck the benthos independent of, -diversity (a common issue in macroecological
with its jaw open. studies). Local -diversity was only moderately correlated with
We ignored all observations of juvenile fishes <10 cm, as they the richness component of LCBD (r = −0.55) and showed little cor-
could not be accurately identified to species. We also ignored the relation with the replacement component of LCBD (r = −0.06),
feeding activities of the sea urchin D. antillarum, as it was not indicating that having all three in the model did not introduce
observed in any of the videos and we later accounted for its effects strong collinearity.
in our models (using urchin density from our benthic survey;
see above). We also noted 25 instances where damselfish (family Statistical analysis
Pomacentridae) interfered with grazing. We removed these feeding All analyses were conducted in R version 3.4.3 (see supplementary
events from our analysis. We removed a further 213 data points code) (40). We initially fit a generalized linear mixed-effects model
where fish spent <4 s in the plot and did not feed, and thus were predicting mass-standardized bite rate using total herbivore bio-
obviously only passing through the plot and not interested in feeding mass, herbivore -diversity, the LCBD reflecting both components

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 5 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

of -diversity, percent coral cover, percent algal turf cover, percent Fig. S2. Accumulation curves for the number of bites observed in each video assay.
Fig. S3. Species accumulation curves for each video assay.
macroalgal cover, and the abundance of D. antillarum as fixed ef-
Table S1. Output from a linear mixed-effects model predicting mass-standardized bite rate
fects. As our random structure, we nested camera within reef within (kg bites per m2 per hour) on benthic turf algae.
location to account for potential nonindependence of observations, Data file S1. Metadata information.
and to account for the impact of unmeasured factors on herbivory Data file S2. Benthic community structure data.
rate. Subsequent exploration of variance inflation factors (VIFs) Data file S3. Herbivore identity, biomass, richness, and bite rate data.
Data file S4. R code script for reproducing all analyses.
revealed strong multicollinearity between coral cover, macroalgal
cover, and turf cover (VIF ≫ 2). We therefore removed macroalgal REFERENCES AND NOTES
cover from the model, focusing on coral cover (which concentrates 1. R. H. Whittaker, Vegetation of the Siskiyou mountains, Oregon and California.
or diffuses herbivore grazing) and algal turf cover (a direct indicator Ecol. Monogr. 30, 279–338 (1960).
of resource availability to grazers). We additionally tested for the 2. B. J. Cardinale, J. E. Duffy, A. Gonzalez, D. U. Hooper, C. Perrings, P. Venail, A. Narwani,
G. M. Mace, D. Tilman, D. A. Wardle, A. P. Kinzig, G. C. Daily, M. Loreau, J. B. Grace,
presence of significant interactions between -diversity and either the A. Larigauderie, D. S. Srivastava, S. Naeem, Biodiversity loss and its impact on humanity.
richness or replacement components of -diversity, but in all cases, Nature 486, 59–67 (2012).
the interaction was not significant (P = 0.20, 0.10, 0.39, and 0.88 for 3. J. E. Duffy, C. M. Godwin, B. J. Cardinale, Biodiversity effects in the wild are common and
richness, replacement, and both interactions in the same model, re- as strong as key drivers of productivity. Nature 549, 261–264 (2017).
4. D. S. Srivastava, M. Vellend, Biodiversity-ecosystem function research: Is it relevant to
spectively). Thus, we chose to present only the main effects of diversity.
conservation? Annu. Rev. Ecol. Evol. Syst. 36, 267–294 (2005).
Our data met the assumptions of normality and equal variance for 5. M. Loreau, N. Mouquet, A. Gonzalez, Biodiversity as spatial insurance in heterogeneous
the above tests, as indicated by graphical analysis of residuals. landscapes. Proc. Natl. Acad. Sci. U.S.A. 100, 12765–12770 (2003).

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

To test whether the effect of species richness in our model was 6. S. Wang, M. Loreau, Biodiversity and ecosystem stability across scales in
due to the presence of one or a few strongly interacting species metacommunities. Ecol. Lett. 19, 510–518 (2016).
7. K. E. France, J. E. Duffy, Diversity and dispersal interactively affect predictability of
versus many species, we fit a general linear model regressing the ecosystem function. Nature 441, 1139–1143 (2006).
proportional biomass of each species against the corresponding 8. P. A. Venail, R. C. MacLean, T. Bouvier, M. A. Brockhurst, M. E. Hochberg, N. Mouquet,
community-wide mass-standardized bite rate. We further investi- Diversity and productivity peak at intermediate dispersal rate in evolving
gated the potential for synergies by including an “average interac- metacommunities. Nature 452, 210–214 (2008).
9. P. Staddon, Z. Lindo, P. D. Crittenden, F. Gilbert, A. Gonzalez, Connectivity, non-random
tion” term, calculated by taking the product of the relative biomass
extinction and ecosystem function in experimental metacommunities. Ecol. Lett. 13,
of each pair of species, and summing across all pairs present in a 543–552 (2010).
community (25). This approach limits us to having to evaluate each 10. J. R. Pasari, T. Levi, E. S. Zavaleta, D. Tilman, Several scales of biodiversity affect ecosystem
pairwise interaction, which is prohibitive at low sample sizes. multifunctionality. Proc. Natl. Acad. Sci. U.S.A. 110, 10219–10222 (2013).
However, this construction assumes that functioning is maximized 11. Y. Hautier, F. Isbell, E. T. Borer, E. W. Seabloom, W. S. Harpole, E. M. Lind, A. S. MacDougall,
C. J. Stevens, P. B. Adler, J. Alberti, J. D. Bakker, L. A. Brudvig, Y. M. Buckley, M. Cadotte,
when all species are even in terms of their biomass, and thus, the M. C. Caldeira, E. J. Chaneton, C. Chu, P. Daleo, C. R. Dickman, J. M. Dwyer, A. Eskelinen,
lack of a significant interaction may not reflect a true lack of synergy, P. A. Fay, J. Firn, N. Hagenah, H. Hillebrand, O. Iribarne, K. P. Kirkman, J. M. H. Knops,
but instead this simplifying assumption (25). The presence of K. J. la Pierre, R. L. McCulley, J. W. Morgan, M. Pärtel, J. Pascual, J. N. Price, S. M. Prober,
significant main effects (i.e., species’ identities) captures significant A. C. Risch, M. Sankaran, M. Schuetz, R. J. Standish, R. Virtanen, G. M. Wardle, L. Yahdjian,
A. Hector, Local loss and spatial homogenization of plant diversity reduce ecosystem
and independent contributions to functioning, and thus may indi-
multifunctionality. Nat. Ecol. Evol. 2, 50–56 (2018).
cate complementarity among multiple species or dominance by one 12. F. van der Plas, P. Manning, S. Soliveres, E. Allan, M. Scherer-Lorenzen, K. Verheyen,
of few species, depending on the community composition. We C. Wirth, M. A. Zavala, E. Ampoorter, L. Baeten, L. Barbaro, J. Bauhus, R. Benavides,
log10-transformed both the response and the average interaction A. Benneter, D. Bonal, O. Bouriaud, H. Bruelheide, F. Bussotti, M. Carnol, B. Castagneyrol,
term not only to satisfy the assumptions of constant variance and Y. Charbonnier, D. A. Coomes, A. Coppi, C. C. Bastias, S. M. Dawud, H. de Wandeler,
T. Domisch, L. Finér, A. Gessler, A. Granier, C. Grossiord, V. Guyot, S. Hättenschwiler,
normality of errors but also to approximate any potentially nonlinear H. Jactel, B. Jaroszewicz, F.-x. Joly, T. Jucker, J. Koricheva, H. Milligan, S. Mueller, B. Muys,
(e.g., power) relationships (28). We also fixed the intercept to 0, as D. Nguyen, M. Pollastrini, S. Ratcliffe, K. Raulund-Rasmussen, F. Selvi, J. Stenlid,
herbivory cannot occur in the absence of all species (25). F. Valladares, L. Vesterdal, D. Zielínski, M. Fischer, Biotic homogenization can decrease
Last, to test the degree to which herbivory rate predicts biologi- landscape-scale forest multifunctionality. Proc. Natl. Acad. Sci. U.S.A. 113, 3557–3562 (2016).
13. A. D. Barnes, P. Weigelt, M. Jochum, D. Ott, D. Hodapp, N. F. Haneda, U. Brose, Species
cally related aspects of reef community structure, we averaged
richness and biomass explain spatial turnover in ecosystem functioning across tropical
mass-standardized bite rate per reef (n = 10) and used it to predict and temperate ecosystems. Philos. Trans. R. Soc. B Biol. Sci. 371, 20150279 (2016).
average algal turf canopy height at the reef scale. We included two 14. H. M. Burley, K. Mokany, S. Ferrier, S. W. Laffan, K. J. Williams, T. D. Harwood, Primary
additional covariates: coral cover (which concentrates or diffuses her- productivity is weakly related to floristic alpha and beta diversity across Australia.
bivore grazing) and the abundance of the sea urchin D. antillarum Glob. Ecol. Biogeogr. 25, 1294–1307 (2016).
15. T. P. Hughes, N. A. J. Graham, J. B. C. Jackson, P. J. Mumby, R. S. Steneck, Rising to the
[which strongly reduces algal cover when at high densities, and thus challenge of sustaining coral reef resilience. Trends Ecol. Evol. 25, 633–642 (2010).
directly competes with herbivorous fishes for algal resources when 16. T. P. Hughes, M. L. Barnes, D. R. Bellwood, J. E. Cinner, G. S. Cumming, J. B. C. Jackson,
at high density; (24)]. We further predicted the effect of algal turf J. Kleypas, I. A. van de Leemput, J. M. Lough, T. H. Morrison, S. R. Palumbi, E. H. van Nes,
canopy height on the density of juvenile corals—given its known M. Scheffer, Coral reefs in the Anthropocene. Nature 546, 82–90 (2017).
17. D. E. Burkepile, M. E. Hay, Herbivore species richness and feeding complementarity affect
negative impact on coral recruitment (30)—using simple linear re-
community structure and function on a coral reef. Proc. Natl. Acad. Sci. U.S.A. 105,
gression. Both responses were log10-transformed to meet assump- 16201–16206 (2008).
tions of constant variance and normality of errors. 18. D. E. Burkepile, M. E. Hay, Feeding complementarity versus redundancy among
herbivorous fishes on a Caribbean reef. Coral Reefs 30, 351–362 (2011).
19. D. B. Rasher, A. S. Hoey, M. E. Hay, Consumer diversity interacts with prey defenses to
SUPPLEMENTARY MATERIALS drive ecosystem function. Ecology 94, 1347–1358 (2013).
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ 20. A. S. Hoey, D. R. Bellwood, Limited functional redundancy in a high diversity system:
content/full/5/3/eaav6420/DC1 Single species dominates key ecological process on coral reefs. Ecosystems 12,
Fig. S1. Locations of our 10 study sites distributed across the Dominican Republic. 1316–1328 (2009).

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 6 of 7

SCIENCE ADVANCES | RESEARCH ARTICLE

21. B. J. Cardinale, A. R. Ives, P. Inchausti, Effects of species diversity on the primary 35. M. Dornelas, N. J. Gotelli, B. McGill, H. Shimadzu, F. Moyes, C. Sievers, A. E. Magurran,
productivity of ecosystems: Extending our spatial and temporal scales of inference. Assemblage time series reveal biodiversity change but not systematic loss. Science 344,
Oikos 104, 437–450 (2004). 296–299 (2014).
22. F. Isbell, J. Cowles, L. E. Dee, M. Loreau, P. B. Reich, A. Gonzalez, A. Hector, B. Schmid, 36. J. P. Hawkins, C. M. Roberts, Effects of artisanal fishing on Caribbean coral reefs.
Quantifying effects of biodiversity on ecosystem functioning across times and places. Conserv. Biol. 18, 215–226 (2004).
Ecol. Lett. 21, 763–778 (2018). 37. E. L. A. Kelly, Y. Eynaud, S. M. Clements, M. Gleason, R. T. Sparks, I. D. Williams, J. E. Smith,
23. P. Legendre, M. De Cáceres, Beta diversity as the variance of community data: Investigating functional redundancy versus complementarity in Hawaiian herbivorous
Dissimilarity coefficients and partitioning. Ecol. Lett. 16, 951–963 (2013). coral reef fishes. Oecologia 182, 1151–1163 (2016).
24. R. C. Carpenter, Mass mortality of a Caribbean sea urchin: Immediate effects on 38. G. Roff, P. J. Mumby, Global disparity in the resilience of coral reefs. Trends Ecol. Evol. 27,
community metabolism and other herbivores. Proc. Natl. Acad. Sci. U.S.A. 85, 511–514 404–413 (2012).
(1988). 39. P. Legendre, Interpreting the replacement and richness difference components of beta
25. L. Kirwan, J. Connolly, J. A. Finn, C. Brophy, A. Lüscher, D. Nyfeler, M. T. Sebastià, diversity. Glob. Ecol. Biogeogr. 23, 1324–1334 (2014).
Diversity-interaction modeling: Estimating contributions of species identities and 40. R Core Team, R: A Language and Environment for Statistical Computing (R Foundation for
interactions to ecosystem function. Ecology 90, 2032–2038 (2009). Statistical Computing, 2017).
26. T. C. Adam, M. Kelley, B. I. Ruttenberg, D. E. Burkepile, Resource partitioning along
multiple niche axes drives functional diversity in parrotfishes on Caribbean coral reefs.
Acknowledgments: This work was conducted under research permissions granted to R.E.T.
Oecologia 179, 1173–1185 (2015).
for ecological monitoring work in 2017. We thank R. de Leon for assistance in the field.
27. S. J. Brandl, D. R. Bellwood, Microtopographic refuges shape consumer-producer
Funding: D.B.R. and J.S.L. were supported by Bigelow Laboratory for Ocean Sciences. Benthic
dynamics by mediating consumer functional diversity. Oecologia 182, 203–217 (2016).
surveys were carried out with funds awarded to R.E.T. and R.S.S. from Propogas DR,
28. J. Connolly, T. Bell, T. Bolger, C. Brophy, T. Carnus, J. A. Finn, L. Kirwan, F. Isbell, J. Levine,
Incorporated. A.A.I.-G. was supported by a National Science Foundation Research Experience
A. Lüscher, V. Picasso, C. Roscher, M. T. Sebastia, M. Suter, A. Weigelt, An improved model
for Undergraduates program at Bigelow Laboratory (NSF EAR 1460861). Author
to predict the effects of changing biodiversity levels on ecosystem function. J. Ecol. 101,
contributions: D.B.R., S.J.B., and J.S.L. conceived the study; D.B.R., R.E.T., and R.S.S. collected

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

344–355 (2013).
the data; A.A.I.-G. analyzed the video footage; J.S.L. conducted the analyses; and J.S.L. and
29. R. Winfree, J. R. Reilly, I. Bartomeus, D. P. Cariveau, N. M. Williams, J. Gibbs, Species
D.B.R. wrote the manuscript with input from all coauthors. Competing interests: The authors
turnover promotes the importance of bee diversity for crop pollination at regional scales.
declare that they have no competing interests. Data and materials availability: All data
Science 359, 791–793 (2018).
needed to evaluate the conclusions in the paper are present in the paper and/or the
30. S. N. Arnold, R. S. Steneck, P. J. Mumby, Running the gauntlet: Inhibitory effects of algal
Supplementary Materials. Additional data related to this paper may be requested from the
turfs on the processes of coral recruitment. Mar. Ecol. Prog. Ser. 414, 91–105 (2010).
authors.
31. R. S. Steneck, S. N. Arnold, P. J. Mumby, Experiment mimics fishing on parrotfish: Insights
on coral reef recovery and alternative attractors. Mar. Ecol. Prog. Ser. 506, 115–127 (2014).
Submitted 6 October 2018
32. R. S. Steneck, P. J. Mumby, C. Macdonald, D. B. Rasher, G. Stoyle, Attenuating effects of
Accepted 23 January 2019
ecosystem management on coral reefs. Sci. Adv. 4, eaao5493 (2018).
Published 6 March 2019
33. Y.-M. Bozec, S. O’Farrell, J. H. Bruggemann, B. E. Luckhurst, P. J. Mumby, Tradeoffs
10.1126/sciadv.aav6420
between fisheries harvest and the resilience of coral reefs. Proc. Natl. Acad. Sci. U.S.A. 113,
4536–4541 (2016).
34. M. Vellend, L. Baeten, I. H. Myers-Smith, S. C. Elmendorf, R. Beauséjour, C. D. Brown, Citation: J. S. Lefcheck, A. A. Innes-Gold, S. J. Brandl, R. S. Steneck, R. E. Torres, D. B. Rasher,
P. De Frenne, K. Verheyen, S. Wipf, Global meta-analysis reveals no net change in local-scale Tropical fish diversity enhances coral reef functioning across multiple scales. Sci. Adv. 5,
plant biodiversity over time. Proc. Natl. Acad. Sci. U.S.A. 110, 19456–19459 (2013). eaav6420 (2019).

Lefcheck et al., Sci. Adv. 2019; 5 : eaav6420 6 March 2019 7 of 7

Tropical fish diversity enhances coral reef functioning across multiple scales
Jonathan S. Lefcheck, Anne A. Innes-Gold, Simon J. Brandl, Robert S. Steneck, Ruben E. Torres and Douglas B. Rasher

Sci Adv 5 (3), eaav6420.

DOI: 10.1126/sciadv.aav6420

Downloaded from http://advances.sciencemag.org/ on August 30, 2019

ARTICLE TOOLS http://advances.sciencemag.org/content/5/3/eaav6420

SUPPLEMENTARY http://advances.sciencemag.org/content/suppl/2019/03/04/5.3.eaav6420.DC1
MATERIALS

REFERENCES This article cites 39 articles, 10 of which you can access for free
http://advances.sciencemag.org/content/5/3/eaav6420#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science Advances (ISSN 2375-2548) is published by the American Association for the Advancement of Science, 1200 New
York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive licensee American
Association for the Advancement of Science. No claim to original U.S. Government Works. The title Science Advances is a
registered trademark of AAAS.