1 Biodiversity at multiple trophic levels is needed for ecosystem multifunctionality
1 2
Santiago Soliveres1,*, Fons van der Plas1,2, Peter Manning1,2, Daniel Prati1, Martin M.
3
Gossner3,4, Swen C. Renner5,6, Fabian Alt7, Hartmut Arndt8, Vanessa Baumgartner9, Julia 4
Binkenstein10, Klaus Birkhofer11, Stefan Blaser1, Nico Blüthgen12, Steffen Boch1,13, Stefan 5
Böhm5, Carmen Börschig14, Francois Buscot15,16, Tim Diekötter17, Johannes Heinze18,19, 6
Norbert Hölzel20, Kirsten Jung21, Valentin H. Klaus20, Till Kleinebecker20, Sandra Klemmer15, 7
Jochen Krauss22, Markus Lange23, E. Kathryn Morris24,25, Jörg Müller18, Yvonne Oelmann7, 8
Jörg Overmann9, Esther Pašalić3, Matthias C. Rillig19,25, H. Martin Schaefer26, Michael 9
Schloter27, Barbara Schmitt1, Ingo Schöning23, Marion Schrumpf23, Johannes Sikorski9, 10
Stephanie A. Socher28, Emily F. Solly23,29, Ilja Sonnemann30, Elisabeth Sorkau7, Juliane 11
Steckel22, Ingolf Steffan-Dewenter22, Barbara Stempfhuber27, Marco Tschapka21,31, Manfred 12
Türke16,32, Paul C. Venter8, Christiane N. Weiner12, Wolfgang W. Weisser3,4, Michael 13
Werner22, Catrin Westphal14, Wolfgang Wilcke33, Volkmar Wolters34, Tesfaye Wubet15,16, 14
Susanne Wurst30, Markus Fischer1,2,13, Eric Allan1,35 15
16
1Institute of Plant Sciences, University of Bern, Altenbergrain 21, 3013 Bern, Switzerland.
17
2Senckenberg Gesellschaft für Naturforschung, Biodiversity and Climate Research Centre 18
BIK-F, Senckenberganlage 25, 60325 Frankfurt, Germany. 3Institute of Ecology, Friedrich- 19
Schiller-University Jena, Dornburger Straße 159, D-07743 Jena, Germany. 4Technische 20
Universität München, Terrestrial Ecology Research Group, Department of Ecology and 21
Ecosystem Management, School of Life Sciences Weihenstephan, Hans-Carl-von-Carlowitz- 22
Platz 2, 85354 Freising, Germany. 5Institute of Zoology, University of Natural Resources and 23
Life Science, Gregor-Mendel-Straße 33, 1180 Vienna, Austria. 6Smithsonian Conservation 24
Biology Institute, National Zoological Park, 1500 Remount Road, Front Royal, VA 22630 25
US. 7Geocology, University of Tuebingen, Ruemelinstr. 19-23, 72070 Tuebingen, Germany.
26
This document is the accepted manuscript version of the following article:
Soliveres, S., van der Plas, F., Manning, P., Prati, D., Gossner, M. M., Renner, S. C., … Allan, E. (2016). Biodiversity at multiple trophic levels is needed for ecosystem multifunctionality. Nature, 536(7617), 456-459.
https://doi.org/10.1038/nature19092
2
8University of Cologne, Institute for Zoology, Zülpicher Str. 47b, 50674 Cologne. 9Leibniz 27
Institute DSMZ-German Collection of Microorganisms and Cell Cultures, Inhoffenstr. 7B, 28
38124 Braunschweig, Germany.10Chair of Nature Conservation and Landscape Ecology, 29
Faculty of Environment and Natural Resources, University of Freiburg, Tennenbacher 30
Straße 4, 79106 Freiburg, Germany 11Department of Biology, Lund University.12Ecological 31
Networks, Biology, Technische Universität Darmstadt, Schnittspahnstr. 3, 64287 Darmstadt.
32
13Botanical Garden, University of Bern, Altenbergrain 21, 3013 Bern, Switzerland.
33
14Agroecology, Department of Crop Sciences, Georg-August University of Göttingen, 34
Grisebachstr. 6, D-37077, Göttingen, Germany. 15UFZ-Helmholtz Centre for Environmental 35
Research, Department of Soil Ecology, Theodor-Lieser-Straße 4, 06120 Halle (Saale), 36
Germany. 16German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, 37
Deutscher Platz 5e, D-04103 Leipzig, Germany. 17Department of Landscape Ecology, Kiel 38
University, Olshausenstr. 75, D-24118 Kiel Germany. 18Biodiversity Research / Systematic 39
Botany, University of Potsdam, Maulbeerallee 1, D-14469 Potsdam, Germany. 19Berlin- 40
Brandenburg Institute of Advanced Biodiversity Research (BBIB), D-14195 Berlin, Germany.
41
20Institute of Landscape Ecology, University of Münster, Heisenbergstr. 2, 48149 Münster, 42
Germany. 21Institute of Evolutionary Ecology and Conservation Genomics, University of 43
Ulm, Albert-Einstein-Allee 11, 89069 Ulm, Germany. 22Department of Animal Ecology and 44
Tropical Biology, Biocentre, University of Würzburg, Am Hubland, D-97074 Würzburg, 45
Germany. 23Max-Planck Institute for Biogeochemistry, Hans-Knoell-Str. 10, 07745 Jena, 46
Germany. 24Xavier University, Department of Biology, 3800 Victory Parkway, Cincinnati, 47
OH 45207. 25Plant Ecology, Institut für Biologie, Freie Universität Berlin, Altensteinstr. 6, D- 48
14195 Berlin. 26Department of Ecology and Evolutionary Biology, Faculty of Biology, 49
University of Freiburg, Hauptstraße 1, 79104 Freiburg i. Br., Germany. 27Research Unit for 50
Environmental Genomics; Helmholtz Zentrum München, Ingolstädter Landstr. 1, 85758 51
Oberschleissheim, Germany. 28Department of Ecology and Evolution, Universität Salzburg, 52
3 Hellbrunnerstrasse 34, 5020 Salzburg, Austria. 29Swiss Federal Institute for Forest, Snow and 53
Landscape Research WSL, Zürcherstrasse 111, 8903 Birmensdorf, Switzerland. 30Functional 54
Biodiversity, Institute of Biology, Freie Universität Berlin. Königin-Luise-Str. 1-3. D-14195 55
Berlin. 31Smithsonian Tropical Research Institute, Balboa, Panama. 32Institute for Biology, 56
Leipzig University, Johannisallee 21, D-04103 Leipzig, Germany. 33Institute of Geography 57
and Geoecology, Karlsruhe Institute of Technology (KIT), Reinhard-Baumeister-Platz 1, 58
76131 Karlsruhe, Germany. 34Department of Animal Ecology, Justus Liebig University 59
Giessen. 35Centre for Development and Environment, University of Bern, Hallerstrasse, 10, 60
3012 Bern, Switzerland.
61 62
*corresponding author: santiago.soliveres@ips.unibe.ch 63
4 Abstract
64
Many experiments have shown that biodiversity loss reduces the capacity of ecosystems 65
to provide the multiple services on which humans depend1,2. However, experiments 66
necessarily simplify the complexity of natural ecosystems and normally control for other 67
important drivers of ecosystem functioning such as the environment or land use.
68
Additionally, existing studies typically focus on the diversity of single trophic groups, 69
neglecting the fact that biodiversity loss occurs across many taxa3,4, and that the 70
functional effects of any trophic group may depend on the abundance and diversity of 71
others5,6. We analysed the relationships between the species richness and abundance of 72
nine trophic groups, including 4,600 above- and belowground taxa, with 14 ecosystem 73
services and functions (hereafter services) and with their simultaneous provision 74
(multifunctionality) in 150 grasslands. Here we show that high richness of multiple 75
trophic groups (multitrophic richness) had stronger positive effects on ecosystem 76
services than any of them individually, including plant species richness (the most widely 77
used measure of biodiversity). Three trophic groups, on average, influenced each 78
ecosystem service, and each trophic group influenced at least one service. Multitrophic 79
richness was particularly beneficial regulating and cultural services, and for 80
multifunctionality, whereas changes in the total abundance of multiple trophic groups 81
(multitrophic abundance) positively affected supporting services. Multitrophic richness 82
and abundance were as strong drivers of ecosystem functioning as abiotic conditions and 83
land-use intensity, extending previous experimental results7,8 to real-world ecosystems.
84
Primary producers, herbivorous insects and microbial decomposers seem to be 85
particularly important drivers of ecosystem functioning due to strong and frequent 86
positive associations of their richness or abundance with multiple ecosystem services.
87
Our results show that multitrophic richness and abundance support ecosystem 88
5 functioning, and demonstrate that a focus on single groups have greatly underestimated 89
the functional importance of biodiversity.
90 91
MAINTEXT 92
Global change is causing species losses across many trophic groups3-4, with potential effects 93
on the services that ecosystems provide to humans1-2. The functional consequences of 94
biodiversity declines across multiple trophic groups are hard to predict from studies focusing 95
on single taxa, as the functional effects of different groups may complement or oppose each 96
other5,6,9,10. For example, diversity effects of plants and microbes are complementary, 97
maximising rates of nutrient cycling11; while plant and herbivore diversity have opposing 98
effects on biomass stocks10,12-13. Consequently, we know very little about the relative effect of 99
changes in the diversity of different trophic groups on the provision of individual2,5,6,9,13,14 or 100
multiple (multifunctionality)11,15 ecosystem services.
101
In addition to decreasing species richness, global change is altering the total 102
abundance (total number of individuals, or biomass, within communities) of multiple trophic 103
groups4. Changes in abundance could mitigate or exacerbate the functional consequences of 104
species loss16,17 by influencing the ability of each trophic group to capture resources.
105
However, studies normally focus on the effects of community evenness or of dominant 106
species18-20, whereas the simultaneous effects of changes in richness and total abundance on 107
ecosystem functioning are largely unexplored6,16. The relative importance of richness and 108
abundance may depend on the function or service of interest; total abundance could be a main 109
driver of biogeochemical processes rates (e.g., biomass production18, nutrient capture and 110
cycling). In contrast, ecosystem services related to biotic interactions (e.g., pollination, pest 111
control) could be predominantly driven by species richness16. Ecosystem services also depend 112
on abiotic factors and, although experiments show that effects of biodiversity loss on 113
6 functioning are as large as those of abiotic drivers7,8, it is unclear whether species richness and 114
abundance are similarly important in real-world ecosystems6,14,21,24. 115
Here we adopted a multitrophic approach to evaluate biodiversity-multifunctionality 116
relationships in 150 real-world grasslands. We measured the richness and abundance of nine 117
trophic groups: primary producers, above- and belowground herbivores and predators, 118
detritivores, soil microbial decomposers, plant symbionts, and bacterivores. These trophic 119
groups comprised 4,600 plant, animal and microbial taxa, and were measured alongside 14 120
ecosystem variables (both functions and service proxies; referred to as services hereafter).
121
These are related to the four main types of ecosystem services23: provisioning (fodder 122
production and quality), supporting (potential nitrification, P retention, root biomass and 123
decomposition rate, mycorrhizal colonization and soil aggregate stability), regulating (soil 124
carbon, pollinator abundance, pest control, resistance to pathogens) and cultural services 125
(recreation benefits of flower cover and bird diversity). We fitted linear models to test for 126
both positive and negative relationships between the species richness and abundance within 127
the nine trophic groups and each ecosystem service, the four types of services (provisioning, 128
supporting, regulating and cultural), and multifunctionality24 (see Methods). We accounted 129
for potential confounding factors by performing our analyses on residuals, after controlling 130
for land-use intensity, soils and climate. We compared our results with models including only 131
plant species richness, the most commonly used measure of biodiversity21,22,25, and with 132
models including the richness and abundance of each individual trophic group. Additional 133
analyses compared the amount of variance explained by, and the effect size (standardised 134
slope) of, richness and abundance with those of land-use intensity and environmental 135
variables.
136
Individual ecosystem services, service types, and multifunctionality were better 137
predicted by multitrophic richness and abundance than by the richness or abundance of any 138
individual trophic group (Fig. 1; Extended data Fig. 1). The most parsimonious models 139
7 included the richness and/or abundance of 3.14±0.36 trophic groups (average±SE across all 140
14 services). These results remained when using raw data instead of environment-corrected 141
residuals (Extended data Fig. 2), different combinations of ecosystem services (Extended data 142
Fig. 3), and even when we accounted for well-established links between predictor and service 143
in our models (e.g., plant cover vs. biomass). Multitrophic richness (the combined effect of 144
high species richness of multiple trophic groups) had stronger and more positive relationships 145
with provisioning, regulating and cultural services than plant richness alone (Fig. 1, Extended 146
data Fig. 1). For example, both plant and predator richness were related to high levels of pest 147
control, suggesting that combined top-down and bottom-up effectsof diversity26 maximise the 148
provision of this regulating service. Multitrophic richness also had a more positive effect than 149
the strongest positive richness effect found across all individual trophic groups on regulating 150
and cultural services. The findings of our observational study were supported by a 151
quantitative review of the few studies that manipulated the richness of more than one group.
152
Our review showed that including the richness of a second trophic group increased the 153
variance in ecosystem functioning explained by 14-96% for litter decomposition14, biomass 154
production2,12,26, or the number of carbon sources used5 (Extended data Table 1). Collectively, 155
our results show that high species richness of multiple trophic groups is necessary to maintain 156
high levels of ecosystem functioning, particularly for regulating and cultural services.
157
Alongside multitrophic richness, the combined effect of high total abundance of 158
multiple trophic groups (multitrophic abundance), strongly affected ecosystem functioning 159
(according to the amount of variance explained and its effect size). Multitrophic abundance 160
had positive relationships with provisioning and supporting services, but these were generally 161
weaker than those found for the individual trophic group having the strongest positive effect.
162
This suggests that abundance of different trophic groups dampen each other's effects on 163
ecosystem functioning (e.g., higher abundance of predators partially counteracted the positive 164
effects of abundant herbivores on supporting services; Fig. 1). These contrasting effects 165
8 caused multitrophic abundance to increase multifunctionality only at low to moderate levels 166
(Extended data Fig. 1). Overall, our results support the important role of species richness1-2,14- 167
17, 22, 25 as a driver of ecosystem functioning, while also highlighting the overlooked effect of 168
total abundance on supporting and provisioning services. Our study also suggests that the 169
abundances of different trophic groups tend to dampen each other's effects while richness of 170
different groups generally complement each other's positive effects on ecosystem services.
171
To test how general were the multitrophic richness and abundance-functioning 172
relationships we found, we calculated multifunctionality metrics produced using all possible 173
combinations of services. High richness or abundance of multiple trophic groups had 174
increasingly positive effects as more services were considered, and this was consistent across 175
a wide range of levels of multifunctionality (Extended data Fig. 3). To further explore this 176
result, we calculated how similar were the identities of the trophic groups driving a given pair 177
of ecosystem services (functional overlap [ō]25). On average, we found functional overlaps 178
lower than 30% (ō = 0.27±0.03, mean±SE); similar to results found for plant species in 179
grassland experiments (range in ō = 0.19-0.49)25. This demonstrates low multitrophic 180
redundancy and means that different services are supported by different trophic groups (Fig.
181
1; Extended data Fig. 1). We also found that different groups positively affected 182
multifunctionality when it was calculated according to scenarios representing different land- 183
use objectives (Extended data Fig. 4). Finally, five of the nine trophic groups had net positive 184
effects on at least one ecosystem service (e.g., primary producers for pest control, soil 185
microbial decomposers for aggregate stability; Extended data Fig. 1), and all of them affected 186
at least one service. These results collectively show the low functional redundancy found 187
between the multiple trophic groups studied and explains why the richness of more trophic 188
groups is needed to support multifunctionality required at high levels or including more 189
services.
190
9 The strength and direction of the relationships between richness, abundance and 191
ecosystem services were not always positive (Figs. 1 and 2; Extended data Fig. 1), as found 192
by previous studies27,28. Negative relationships might be explained by interference between 193
species within a given trophic group10, or compositional shifts leading to declines in 194
ecosystem functioning17,20,29. Despite these negative associations or potential disservices, our 195
results suggest that, among the many trophic groups studied, those particularly important to 196
maintain the set of services considered are aboveground herbivorous insects, primary 197
producers and soil microbial decomposers. The richness or abundance of these groups were 198
most often related to multifunctionality (43-72% of the 501 possible combinations between 199
the services we measured), and had net positive effects across all services (Fig. 2). These 200
three groups also showed strong and frequent positive associations with the four main 201
ecosystem service types (Fig. 1). These results agree with other studies that have identified 202
plants and soil microorganisms as key drivers of ecosystem functioning11,14,15,29, extending 203
these findings to the richness and abundance of different trophic groups, including primary 204
producers and consumers both above- and belowground. The richness of some of these 205
functionally important trophic groups relate to whole ecosystem diversity3,30 and, thus, 206
management strategies focused on them may foster synergies between biodiversity 207
conservation and high multifunctionality levels.
208
The relative importance of multitrophic richness and abundance compared to 209
environmental drivers of ecosystem functioning has been rarely studied outside experiments 210
or individual functions7-8,11,15,21,22. We therefore calculated the proportion of variance in 211
ecosystem functioning explained by multitrophic richness, abundance and environmental 212
(soil, topography and land-use) factors. Our models accounted for a large proportion (54- 213
64%) of the variance in provisioning, supporting, regulating and cultural ecosystem service 214
types (Fig. 3; Extended data Fig. 5). Multitrophic richness and abundance explained as much 215
variance in ecosystem functioning as abiotic conditions or land-use intensity did, and 216
10 generally had stronger effects (Fig. 3; Extended Data Fig. 5). These results provide evidence 217
that biodiversity is an important driver of ecosystem functioning in comparison with 218
environmental factors, not only for individual functions in small-scale experiments7,8, but also 219
for multiple ecosystem services in realistic landscapes (see also refs. 15, 22).
220
Our study shows that we have greatly underestimated the functional importance of 221
biodiversity in real-world ecosystems by focusing on individual trophic groups, and 222
demonstrate that the functional effects of multitrophic richness and abundance are as strong, 223
or even stronger, than those of the environment or land-use intensity. We identified primary 224
producers, aboveground herbivores and soil decomposers as particularly important trophic 225
groups to maintain ecosystem functioning. Our results suggests that is important to preserve 226
high levels of richness and/or abundance within a wide range of taxa, including taxa often 227
ignored by conservation, such as soil microbial decomposers15, or considered pests in 228
agricultural systems, such as herbivorous insects, if we are to promote high levels of the 229
multiple ecosystem services upon which human well-being depends.
230 231
References 232
1. Cardinale, B.J. et al. Biodiversity loss and its impact on humanity. Nature 486, 59–
233
67 (2012).
234
2. Naeem, S., Duffy, J.E. & Zavaleta E. The Functions of Biological Diversity in an 235
Age of Extinction. Science 336, 1401–1406 (2012).
236
3. Allan, E. et al. Interannual variation in land-use intensity enhances grassland 237
multidiversity. Proc. Natl Acad. Sci. USA 111, 308–313 (2014).
238
4. Newbold, T. et al. Global effects of land use on local terrestrial biodiversity. Nature 239
520, 45–50 (2015).
240
5. Naeem, S., Hahn D. R. & Schuurman, G. Producer-decomposer co-dependency 241
influences biodiversity effects. Nature 403, 762-764. (2000).
242
11 6. Balvanera, P. et al. Linking Biodiversity and Ecosystem Services: Current
243
Uncertainties and the Necessary Next Steps. BioScience 64: 49-57 (2014) 244
7. Hooper, D. U. et al. A global synthesis reveals biodiversity loss as a major driver of 245
ecosystem change. Nature 486, 105–108 (2012).
246
8. Tilman, D., Reich, P. B. & Isbell, F. Biodiversity impacts ecosystem productivity as 247
much as resources, disturbance, or herbivory. Proc. Natl Acad. Sci. USA 109, 248
10394–10397 (2012).
249
9. Petchey, O.L., McPhearson, P.T., Casey, T.M. & Morin, P.J. Environmental 250
warming alters food-web structure and ecosystem function. Nature 402, 69-72 251
(1999).
252
10. Duffy, J. E. et al. The functional role of biodiversity in ecosystems: incorporating 253
trophic complexity. Ecol. Lett. 10, 522–538 (2007).
254
11. Jing, X. et al. The links between ecosystem multifunctionality and above- and 255
belowground biodiversity are mediated by climate. Nat. Commun. 6, 8159 (2015).
256
12. Douglass, J. G., Duffy, J. E. & Bruno, J. F. Herbivore and predator diversity 257
interactively affect ecosystem properties in an experimental marine community.
258
Ecol. Lett. 11, 598–608 (2008).
259
13. Deraison, H., Badenhausser, I., Loeuille, N., Scherber, C., Gross, N. Functional trait 260
diversity across trophic levels determines herbivore impact on plant community 261
biomass. Ecol. Lett. 18, 1346–1355 (2015).
262
14. Handa, I. T. et al. Consequences of biodiversity loss for litter decomposition across 263
biomes. Nature 509, 218–221 (2014).
264
15. Delgado-Baquerizo, M. et al. Microbial diversity drives multifunctionality in 265
terrestrial ecosystems. Nat. Commun. 7, 10541 (2016).
266
16. Garibaldi L.A. et al. Wild Pollinators Enhance Fruit Set of Crops Regardless of 267
Honey Bee Abundance. Science 339, 1608-1611 (2013).
268
12 17. McGrady-Steed, J. Harry, P. M., Morin, P. J. Biodiversity regulates ecosystem 269
predictability. Nature 390, 162-165 (1997).
270
18. Grime, J.P. Benefits of plant diversity to ecosystems: immediate, filter and founder 271
effects. J. Ecol. 86, 902-910 (1998).
272
19. Hillebrand, H., Bennett, D.M. & Cadotte, M.W.Consequences of dominance: a 273
review of evenness effects on local and regional ecosystem processes.Ecology 89, 274
1510–1520 (2008).
275
20. Soliveres, S., et al. Locally rare species influence grassland ecosystem 276
multifunctionality. Phil. Trans. R. Soc. B 371, 20150269 (2016).
277
21. Grace, J. B., et al. Integrative modelling reveals mechanisms linking productivity 278
and plant species richness. Nature 529, 390–393 (2016).
279
22. Maestre, F. T. et al. Plant species richness and ecosystem multifunctionality in 280
global drylands. Science 335, 214–218 (2012).
281
23. Millennium Ecosystem Assessment. Ecosystems and Human Well-Being 1–86 282
World Resources Institute (2005).
283
24. Byrnes, J. E. K. et al. Investigating the relationship between biodiversity and 284
ecosystem multifunctionality: challenges and solutions. Methods Ecol. Evol 5, 111–
285
124 (2014).
286
25. Hector, A. & Bagchi, R. Biodiversity and ecosystem multifunctionality. Nature 448, 287
188–191 (2007).
288
26. Bruno, J. F., Boyer, K. E., Duffy, E. K. & Lee, S. C. Relative and interactive effects 289
of plant and grazer richness in a benthic marine community. Ecology 89, 2518–2528 290
(2008).
291
27. Balvanera, P. et al. Quantifying the evidence for biodiversity effects on ecosystem 292
functioning and services. Ecol. Lett. 9, 1146–1156 (2006).
293
13 28. Lefcheck, J. S., et al. Biodiversity enhances ecosystem multifunctionality across 294
trophic levels and habitats. Nat. Commun. 6, 6936 (2015).
295
29. Naeem, S., Thompson, L. J., Lawler, S. P., Lawton, J. H. & Woodfin, R. M.
296
Declining biodiversity can alter the performance of ecosystems. Nature 368, 734–
297
737 (1994).
298
30. Manning, P. et al. Grassland management intensification weakens the associations 299
among the diversities of multiple plant and animal taxa. Ecology 96, 1492–1501 300
(2014).
301 302
14 FIGURE LEGENDS
303
Figure 1 | Effects of multitrophic richness and abundance on grassland ecosystem 304
services. A-D) Variance explained (marginal R2) and standardised effects for each ecosystem 305
service when models included abundance and richness of multiple or individual trophic 306
groups, E-H) Standardised effects of richness (full bars) and abundance (hatched bars) of 307
individual trophic groups on each ecosystem service type. Ecosystem services types are 308
represented by plant biomass and forage quality (provisioning); potential nitrification, P 309
retention, mycorrhizal colonization, soil aggregate stability, root biomass and decomposition 310
(supporting); soil carbon, pollinator abundance, pest control, and resistance to pathogens 311
(regulating); flower cover and bird diversity (cultural).
312 313
Figure 2 | Functional importance of multiple trophic groups. A) Proportion of the 314
multifunctionality metrics (calculated using every possible combination of 1-9 services; N = 315
501; see Methods) in which the biotic attributes (richness and/or abundance) of each trophic 316
group was included in the most parsimonious model. B) Functional effects (standardized 317
slopes in the model fitted to all 14 services) of the richness (open bars) and abundance 318
(hatched bars) of each group. Bars are only shown for those predictors included in the most 319
parsimonious models. Icons indicate the different trophic groups and whether they are above- 320
(green) or belowground (brown).
321 322
Figure 3 | Biotic vs abiotic drivers of ecosystem functioning. Variation partitioning for 323
three predictor categories in our statistical models: environment, species richness and total 324
abundance (details in Methods). Diagrams show the average across services within each 325
ecosystem service type (detailed results are in Extended data Figs. 2 and 5). The unique 326
variance explained by each predictor category, the shared variance between these categories 327
(intersections of circles), and the variance not explained by the models (Res.) are shown.
328
15
“Biota” refers to the total variance explained by abundance and richness together.
329
Standardised effect sizes are shown as bar plots.
330 331 332
16 333
334
Figure 1 335
336 337 338 339 340
17 341
Figure 2 342
343
18 344
345 346
Figure 3 347
348
19 METHODS
349
Study sites 350
We selected a total of 150 grassland sites (50 m × 50 m) in three regions of Germany (50 sites 351
per region) to cover a gradient of land-use intensities, characterised by contrasting grazing, 352
fertilisation and mowing levels (www.biodiversity-exploratories.de31). The regions in the 353
south-west (Schwäbische Alb) and the north-east (Schorfheide-Chorin) are UNESCO 354
Biosphere Reserves, whereas the central region is in and around the National Park Hainich.
355
The three regions differ substantially in geology, climate and topography31, covering a range 356
of ~3 °C in mean annual temperature and 500 mm in annual precipitation. Plots in each region 357
cover the range of land-use intensities typical for Central European grasslands. We obtained 358
information on land use via questionnaires sent to land owners, asking about the number and 359
type of livestock (converted to livestock units) and the duration of grazing in each plot, 360
fertilisation (from which we calculated the amount of nitrogen added), and mowing (number 361
of cuts per year31,32). We used this information to calculate three standardized indices 362
summarizing grazing, fertilisation and mowing intensity (see ref. 32 for full methodological 363
details).
364 365
Diversity measures 366
At each site, we measured the species richness and abundance of 9 functional groups using 367
standard methodologies (Extended data Table 2). In total we observed about 4,600 taxa on the 368
150 grasslands studied. The 9 trophic groups were: primary producers (vascular plants and 369
bryophytes), belowground herbivores (herbivorous insect larvae sampled in the soil), 370
belowground predators (carnivorous insect larvae sampled in the soil), detritivores (insects 371
and Diplopoda feeding on litter and other detritus), soil microbial decomposers (soil bacteria), 372
aboveground herbivores (insects feeding solely on aboveground plant material), aboveground 373
predators (carnivorous insects, spiders and Chilopoda), plant symbionts (arbuscular 374
20 mycorrhizal fungi), and bacteria-feeding protists (heterotrophic flagellates and ciliates).
375
Lichens and omnivores were not considered in our analyses as they were too rare. We directly 376
measured species richness for most groups but, richness was quantified as family richness for 377
belowground insects and soil bacteria, and as the number of operational taxonomic units 378
(OTUs) for the mycorrhizae and protists. Abundance of each trophic group was also measured 379
using different methods: number of individuals for arthropods, cover for vascular plants and 380
bryophytes, and relative proportion of sequence reads assigned to each family or OTU 381
(protists, soil bacteria and mycorrhiza). To avoid multicollinearity, we did not include the 382
abundances of protists or detritivores as they were highly correlated (Spearman´s ρ > 0.6) 383
with richness (for more details see Extended data Table 3).
384
We also measured the abundance and richness of foliar fungal pathogens, pollinators 385
and birds; however, to include a broader range of ecosystem services in our analyses, we 386
treated these groups as proxies of ecosystem services. Total pollinator abundance and the 387
inverse of pathogen abundance were treated as proxies of regulating services (pollination and 388
disease regulation), and we used bird species richness as a measure of a cultural ecosystem 389
service. Lepidoptera behave as herbivores during juvenile stages and as pollinators when 390
adults. To avoid accounting for them twice, we only assigned them to one trophic group 391
(pollinators), as the data were counts of the adult butterflies, not the caterpillars.
392 393
Ecosystem functioning measures 394
At each site, we measured 14 different ecosystem variables (both functions and service 395
proxies, hereafter services; Extended data Table 2) and classified them into four types of 396
services following the Millennium Ecosystem Assessment23. These 14 ecosystem services 397
were either i) supporting services related to nutrient capture and cycling (root biomass, root 398
decomposition rates, potential nitrification [based on urease activity in soil samples], 399
phosphorus retention [calculated as a ratio between shoot and microbial P stock and soil 400
21 extractable P], arbuscular mycorrhizal fungal root colonization [measured as hyphal length], 401
soil aggregate stability [proportion of water stable soil aggregates]), ii) provisioning services 402
related to agricultural value (forage production [aboveground plant biomass] and forage 403
quality [based on crude protein and relative forage value]), iii) regulating services for 404
neighbouring crop production or climate regulation (i.e., regulating services: resistance to 405
plant pathogens, pest control, pollinator abundance and soil organic C), or iv) cultural services 406
linked to recreation (bird diversity and flower cover). Because the values for trophic groups 407
and ecosystem functions varied widely, we standardised all variables to a common scale 408
ranging from 0 to 1 according to the following formula: STD = (X-Xmin)/(Xmax-Xmin);
409
where STD is the standardised variable, X, Xmin and Xmax are the target variable, and its 410
minimum and maximum value across all sites, respectively. This made slope estimates for 411
different predictors comparable.
412
We calculated ecosystem multifunctionality metrics from the 14 services as the 413
percentage of measured services (to correct for the fact that some services that had not been 414
measured in all sites) that exceeded a given threshold of their maximum observed level across 415
all study sites. To reduce the influence of outliers we calculated the maximum as the average 416
of the top five sites24,33. Given that any threshold is likely to be arbitrary, the use of multiple 417
thresholds is recommended to better understand the role that biodiversity plays in affecting 418
multifunctionality and to understand trade-offs between functions of interest24. Therefore, we 419
used four different thresholds (25%, 50%, 75% and 90%) to represent a wide spectrum in the 420
analyses performed (Extended data Figs. 1-3). As an alternative approach we also calculated 421
multifunctionality scenarios, weighting the services differently according to different potential 422
stakeholders views (i.e., stakeholders willing only to promote provisioning services vs. other 423
trying to maximise cultural and recreation services or the sustainability of soils and crops;
424
Extended data Fig. 4)34. 425
426
22 Effects of multitrophic richness and abundance on grassland ecosystem services and 427
multifunctionality 428
We used linear models to evaluate the relationships between the species richness and 429
abundance of the nine trophic groups and each of the 14 individual ecosystem services, along 430
with the different multifunctionality metrics (four thresholds and according to different 431
scenarios). In all cases, we used a Gaussian error distribution as the errors of our response 432
variables were normally distributed. We report the effects of the different trophic groups on 433
the different functions as slopes from the multiple regression model, these are corrected for 434
the effects of all other variables in the model. Since our main focus was calculating the 435
independent effects of the richness and abundance of the different trophic groups, we 436
corrected them for co-varying factors. Therefore, we calculated residuals for all our variables 437
(both biotic predictors and functioning measures) from linear models including region, land- 438
use intensity (standardised measures of mowing, grazing and fertilisation intensity) and other 439
important environmental factors (soil type and depth, pH, a topographic wetness index based 440
on position within the slope and orientation, and elevation). As an alternative to using 441
residuals, we also fitted models with all the environmental and land-use predictors 442
(standardised to give comparable coefficients) alongside the diversity and abundance 443
measures and these gave very similar results (Extended Data Fig. 2). Standardised coefficients 444
of the functional effects of richness were very similar whether or not we included abundance 445
(ρ = 0.80, P < 0.0001, N = 162; data not shown). We also fitted models with the abundance 446
and richness of only one individual trophic group to compare the results of the best individual 447
trophic group with the multitrophic analyses (Extended data Fig. 1). Finally, we fitted models 448
with only vascular plant species richness as a predictor. The latter is the most common 449
measure of biodiversity7,8,21,22,25,35-38 and we used it to compare our results with those found in 450
previous studies on biodiversity-ecosystem functioning relationships.
451
23 We performed model simplification using the stepAIC function in R, and further 452
simplified the minimal models produced using AIC by removing all terms that were not 453
significant according to F-ratio tests (Extended data Table 4). We also tried multi-model 454
averaging results (weighting the standardized coefficients of multiple plausible models by 455
their relative AIC weight, but this rendered less conservative results than the ones presented 456
here and, thus, we discarded them. We did not fit interactions between the richness and 457
abundance of different trophic groups, or between those and environmental factors, as this 458
would require a large number of coefficients, would be difficult to interpret and would require 459
an even larger dataset than ours (see ref. 20 for a study evaluating the interaction between 460
land-use and diversity). We did not find evidence of non-linear relationships between our 461
predictors and the ecosystem services measured when checking all bivariate relationships;
462
thus we did not include non-linear terms in the models to keep them simple.
463
Not all trophic groups or ecosystem services were measured on all sites; thus different 464
services were analysed using different sized datasets (N ranged between 111 and 54, 465
depending on the service). The different sampling sizes used were not related to the number 466
of trophic groups included in the most parsimonious model (Spearman´s rank correlation 467
coefficient ρ = 0.32), the increase in variance explained regarding vascular plant species 468
richness (ρ = -0.21) or the net effect of richness or abundance (ρ = -0.01 or 0.05, respectively;
469
N = 14 and P > 0.25 in all cases). Thus fitting models differing in sample size for different 470
services did not affect our results.
471
The inclusion of many predictors in statistical models increases the chance of type I 472
error (false positives). To account for this we used a Bernoulli process to detect false 473
discovery rates, where the probability (p) of finding a given number of significant predictors 474
(K) just by chance is a proportion of the total number of predictors tested (N; 16 in our case:
475
the abundance and richness of 7 and 9 trophic groups, respectively) and the P-value 476
considered significant (α; 0.05 in our case)39,40. The probability of finding three significant 477
24 predictors, on average, as we did, is therefore, p = [16!/(16-3)!3!] × 0.053(1-0.05)16-3 = 478
0.0359, indicating that the effects we found are very unlikely to be spurious. The probability 479
of false discovery rates when considering all models and predictors fit (14 ecosystem services 480
× 16 richness and abundance metrics) and the ones that were significant amongst them (52: 25 481
significant abundance predictors and 27 significant richness predictors) was even lower (p <
482
0.0001). All analyses were performed using R version 3.0.241. 483
484
Net functional effects of the different trophic groups across ecosystem service types 485
We calculated the net effect of each trophic group on each ecosystem service type 486
(provisioning, supporting, regulating and cultural) by fitting all services belonging to these 487
types into a single model. To do so, we added two extra predictors to our models: “service 488
identity” as a fixed factor, to account for differences between individual services, and “site”
489
as a random factor, to account for correlations between services, abundance and richness 490
values measured on the same site. Since we were interested in the net effects of each group 491
across all services, we did not fit interactions between our multitrophic predictors and service 492
identity. The net effect across all services was analysed using the same approach, but fitting a 493
single model for the 14 ecosystem services at the same time. This approach corrects for the 494
fact that the individual service models vary in their explanatory power and in the predictor 495
variables included. Fitting all services into a single model allows us to obtain a robust 496
estimate of the net functional effect (the standardised coefficient from the model) of the 497
abundance and richness of each trophic group on each ecosystems service type and on 498
multifunctionality, together with an estimate of its error. If the standardised coefficient was 499
positive, we interpreted it as a net overall positive effect of either richness or abundance 500
across all services, or on a given service type (Figs. 1 and 2). In all cases, we used 501
standardised coefficients of the most parsimonious models after model reduction. However, 502
our results remained when using other approaches that account for differences in model fit, 503
25 such as multi-model averaging coefficients (coefficients were weighted according to the AIC 504
weight of the models in which each predictor is included) or when weighting the standardised 505
coefficient for each ecosystem service by the adjusted R2 of each model (which should also be 506
comparable across models with different response variables; Extended data Fig. 6).
507 508
Variance partitioning analyses 509
Variance partitioning analyses (also known as commonality analyses) were performed with 510
standard techniques42,43 based on the comparison of variance explained by models including 511
every possible combination of variables. Variables were organized by environment (study 512
region, soil type, pH, topographic wetness index, grazing and fertilisation, with the remaining 513
environmental predictors removed to prevent multicollinearity; Extended data Table 3), 514
species richness (standardised species richness of the 9 trophic groups) and abundance 515
(standardised abundance of those trophic groups in which abundance and richness were not 516
strongly correlated [ρ < 0.6; Extended data Table 3]). Thus, we fitted a series of seven models 517
for each service and multifunctionality metric (at the 25, 50, 75 and 90% thresholds) to 518
extract the unique and shared variance for each combination of variables (environment only, 519
richness only, abundance only, environment + abundance, environment + richness, richness + 520
abundance, and all predictors together). Variance partitioning analyses were performed with 521
the full models (without model simplification) to allow us to compare between different 522
services. As a consequence, we used R2 rather than the adjusted R2 because, due to the large 523
number of predictors, some adjusted R2 values were negative, complicating the extraction of 524
unique variance explained by each predictor. Venn diagrams were drawn using euler APE for 525
Windows44. 526
To compare the effect size between richness, abundance and environment on the 527
different ecosystem services and multifunctionality metrics, we summed the standardized 528
coefficients of all predictors from each component (abundance of 5 trophic groups 529
26 [abundance], richness of 9 trophic groups [richness], and pH, fertilisation, grazing, and 530
topographic wetness index [environment]). We excluded study region and soil type when 531
summing effects, as these were categorical predictors and their coefficients were not 532
straightforward to interpret. We performed these calculations for each of the 14 ecosystem 533
services and 4 multifunctionality metrics in isolation (Extended data Fig. 4), and for each 534
ecosystem service type (Fig. 2) by using models containing all the ecosystem services 535
belonging to each type into a single model (again, adding “service identity" and “site” as fixed 536
and random predictors, respectively).
537 538
Analysing every possible combination of ecosystem services 539
Studies on multifunctionality are difficult to compare as they include different measures of 540
ecosystem functioning. To allow us to generalise our results and to test whether multitrophic 541
richness and abundance are more important in supporting higher numbers of services 542
simultaneously, we also calculated multifunctionality indices using every possible 543
combination of the services we measured. We did this after removing those services with 544
more than 20 missing sites, leaving a total of 9 services (501 combinations), as response 545
variables. We calculated multifunctionality at the 25%, 50%, 75% and 90% thresholds for all 546
these combinations (Extended data Fig. 3). We also tested the sensitivity of our analyses to 547
missing data by repeating our analyses for every possible combination of 1-13 of the 14 548
measured services (16,368 combinations; results for multifunctionality calculated with all 14 549
ecosystem services are presented in Extended data Figs. 1 and 2). To allow the comparison of 550
models with different services, data gaps were filled with the average value of a given service 551
in a given region, which is a conservative approach. In both cases (combinations of 1-13 or 1- 552
9 functions), the most parsimonious among all the possible models were selected based on 553
their AIC. This avoids inflated type I error caused by fitting a large number of models, as 554
27 model selection was not based on P-values. Results using 9 or 14 functions were qualitatively 555
the same and therefore only the former are shown here.
556 557
Review of multitrophic manipulative approaches 558
Manipulative experiments including as many groups and services as we considered in this 559
study do not yet exist. However, we compared our correlational results with available 560
evidence from experiments manipulating the diversity of more than one trophic group. To do 561
this we performed a bibliographic research in the Web of Knowledge and in Google Scholar 562
using all combinations of the terms "multitrophic" or "trophic groups" + "functioning" or 563
"multifunctionality" or "biomass" or "ecosystem services" or "diversity". We also screened 564
references within available reviews on multitrophic diversity-ecosystem functioning 565
relationships10,27,45. Among the papers found, we selected those which fulfilled the following 566
criteria: 1) it was a manipulative study, 2) it included a range in species richness (not only 567
presence/absence) of, at least, two different trophic groups, 3) it provided enough information 568
to calculate the increase in variance explained by the addition of a second trophic group. Only 569
four studies, including seven ecosystem functions, fulfilled these criteria (Extended data 570
Table 1). Some of these manipulative studies did not include plants, so we calculated the 571
percentage increase in variance explained when comparing a model with the trophic group 572
that had the strongest explanatory power with a model containing two trophic groups. When 573
the same function was measured across several studies (i.e., biomass), we calculated the 574
average increase in variance explained for this variable when another trophic level was added.
575
These results were used to qualitatively compare the limited evidence from multitrophic 576
manipulations with our results.
577 578 579 580
28 References
581
31. Fischer, M., et al. Implementing large-scale and long-term functional biodiversity 582
research: The Biodiversity Exploratories. Basic Appl. Ecol. 6, 473–485 (2010).
583
32. Blüthgen, N., et al. A quantitative index of land-use intensity in grasslands:
584
integrating mowing, grazing and fertilization. Basic Appl. Ecol. 13, 207–220 (2012).
585
33. Zavaleta, E., Pasari, J.R., Hulvey, K.B & Tilman, G.D. et al. Sustaining multiple 586
ecosystem functions in grassland communities requires higher biodiversity. Proc.
587
Natl Acad. Sci. USA 107, 1443-1446 (2010).
588
34. Allan, E. et al. Land use intensification alters ecosystem multifunctionality via loss 589
of biodiversity and changes to functional composition. Ecol. Lett. 18, 834-843 590
(2015).
591
35. Hautier, Y. et al. Eutrophication weakens stabilizing effects of diversity in natural 592
grasslands. Nature 508, 521–525 (2014).
593
36. Gamfeldt, L., Hillebrand, H. & Jonsson, P. R. Multiple functions increase the 594
importance of biodiversity for overall ecosystem functioning. Ecology 89, 1223–
595
1231 (2008).
596
37. Gamfeldt, L. et al. Higher levels of multiple ecosystem services are found in forests 597
with more tree species. Nat. Commun 4, 1340 (2013).
598
38. Cardinale, B. J. et al. Effects of biodiversity on the functioning of trophic groups 599
and ecosystems. Nature 443, 989–992 (2006).
600
39. Moran, M.D. Arguments for rejecting the sequential Bonferroni in ecological 601
studies. Oikos 100, 403–405 (2003).
602
40. Tylianakis, J.M., et al. Resource Heterogeneity Moderates the Biodiversity-Function 603
Relationship in Real World Ecosystems. PloS Biol. 6 (5), e122 (2008).
604
41. R Development Core Team. R: a language and environment for statistical 605
computing (R Foundation for Statistical Computing, 2014).
606
29 42. Borcard, D., Legendre, P., Drapeau, P. Partialling out the spatial component of 607
ecological variation. Ecology 73, 1045-1055 (1992).
608
43. Peres-Neto, P., Legendre, P., Dray, S. & Borcard, D. Variation partitioning of 609
species data matrices: estimation and comparison of fractions. Ecology 87, 2614- 610
2625 (2006).
611
44. Micaleff, L. & Rogers, P. eulerAPE: Drawing Area-Proportional 3-Venn Diagrams 612
Using Ellipses. PloS ONE 9(7): e101717. doi:10.1371/journal.pone.0101717 (2014).
613
45. Worm, B. & Duffy, J. E. Biodiversity, productivity and stability in real food webs.
614
Trends Ecol. Evol. 18, 628-632 (2003).
615 616
Supplementary Information is linked to the online version of the paper at 617
www.nature.com/nature.
618 619
Acknowledgements 620
Three anonymous referees, Bernhard Schmid, Fernando T. Maestre and Sonia Kéfi provided 621
comments that improved a previous version of this manuscript. Werner Ulrich and Nick J.
622
Gotelli provided statistical advice. We thank the people largely contributing to maintain the 623
Biodiversity Exploratories program: A. Hemp, K. Wells, S. Gockel, K. Wiesner and M.
624
Gorke (local management team), S. Pfeiffer and C. Fischer (central office), B. König-Ries and 625
M. Owonibi (central database management), and E. Linsenmair, D. Hessenmöller, J.
626
Nieschulze, E-D. Schulze, and the late E. Kalko for their role in setting up the project. This 627
work was funded by the Deutsche Forschungsgemeinschaft Priority Program 1374 628
“Infrastructure-Biodiversity Exploratories”. Fieldwork permits were given by the responsible 629
state environmental offices of Baden-Württemberg, Thüringen, and Brandenburg (according 630
to §72 BbgNatSchG). Figure icons were created by Rubén D. Manzanedo.
631 632
30 Author contributions: SS and EA conceived the idea of this study, MF initiated the
633
Biodiversity Exploratories project aimed at measuring multiple diversities and functions in 634
the field sites. All authors but SS, EA and FVDP contributed data. SS and FVDP performed 635
the analyses. SS and SR performed the literature search. SS wrote the first draft of the 636
manuscript and all the authors (especially EA, PM, FDVP, MMG and DP) contributed 637
substantially to the revisions.
638 639
Author Information:
640
-Reprints and permissions information is available at www.nature.com/reprints.
641
-The authors have no competing financial interests.
642
-Correspondence and requests for materials should be addressed to 643
santiago.soliveres@ips.unibe.ch 644