Increasing crop heterogeneity enhances multitrophic diversity across agricultural regions Clélia Siramia,b,c,1, Nicolas Grossd,e,f, Aliette Bosem Baillodg,h,2, Colette Bertrandi,j,k,2, Romain Carriéb,l,2, Annika Hassg,2, Laura Henckele,m,n,o,2, Paul Miguete,m,n,p,2, Carole Vuillota,q,2, Audrey Alignieri,r, Jude Girards, Péter Batáryg,t, Yann Cloughg,l, Cyrille Viollea, David Giraltu, Gerard Botau, Isabelle Badenhaussere,m,n,v, Gaëtan Lefebvrew, Bertrand Gauffree,m,n, Aude Vialatteb,c, François Calatayudb,c, Assu Gil-Tenax, Lutz Tischendorfs, Scott Mitchells, Kathryn Lindsays, Romain Georgesk, Samuel Hilairew, Jordi Recasensy,z, Xavier Oriol Solé-Senany,z, Irene Robleñoy,z, Jordi Boschaa, Jose Antonio Barrientosbb, Antonio Ricartecc, Maria Ángeles Marcos-Garciacc, Jesús Miñanodd, Raphaël Matheveta, Annick Gibonb, Jacques Baudryi,r, Gérard Balentb, Brigitte Poulinw, Françoise Burelk,r, Teja Tscharntkeg, Vincent Bretagnollee,m, Gavin Siriwardenaee, Annie Ouinb,c, Lluis Brotonsu,x,ff, Jean-Louis Martina,3, and Lenore Fahrigs,3 a UMR 5175 CEFE, CNRS, Université de Montpellier, Université Paul Valéry Montpellier, EPHE, IRD, F-34293 Montpellier, France; bUMR 1201 DYNAFOR, Université de Toulouse, INRA, F-31326 Castanet-Tolosan, France; cLTSER Zone Atelier Pyrénées Garonne, F-31320 Auzeville-Tolosane, France; dUMR Ecosystème Prairial, UCA, INRA, VetAgro Sup, F-63000 Clermont-Ferrand, France; eLTSER Zone Atelier Plaine et Val de Sèvre, F-79360 Villiers en Bois, France; fDepartamento de Biología y Geología, Física y Química Inorgánica, Escuela Superior de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, 28933 Móstoles, Spain; g Agroecology, Department for Crop Sciences, University of Goettingen, D-37077 Göttingen, Germany; hAgroecology and Biodiversity, Research Institute of Organic Agriculture, CH-5070 Frick, Switzerland; iUMR 0980 BAGAP, INRA, Agrocampus Ouest, ESA, F-35042 Rennes, France; jUMR 1402 ECOSYS, INRA, AgroParisTech, Université Paris-Saclay, F-78026 Versailles, France; kUMR 6553 ECOBIO, CNRS, Université de Rennes 1, F-35042 Rennes, France; lCentre for Environmental and Climate Research, Lund University, SE-22362 Lund, Sweden; mUMR 7372 CEBC, CNRS, Université de La Rochelle, F-79360 Villiers en Bois, France; nUSC1339 SEC-LR, INRA, Villiers en Bois, F-79360 Niort, France; oSwedish Species Information Centre, Swedish University of Agricultural Sciences, SE-75007 Uppsala, Sweden; pUR1115 PSH, INRA, F-84914 Avignon, France; qUMR 7204 CESCO, Sorbonne Universités, CNRS, UPMC, Muséum National d’Histoire Naturelle, F-75005 Paris, France; rLTSER, Zone Atelier Armorique, F-35042 Rennes Cedex, France; sGeomatics and Landscape Ecology Laboratory, Carleton University, Ottawa, ON K1S 5B6, Canada; t“Lendület” Landscape and Conservation Ecology, Institute of Ecology and Botany, MTA Centre for Ecological Research, 2163 Vácrátót, Hungary; uCTFC Forest Sciences Centre of Catalonia, 25280 Solsona, Spain; vURP3F, INRA, 86600 Lusignan, France; wTour du Valat, Research Institute for the Conservation of Mediterranean Wetlands, F-13200 Arles, France; xInForest Joint Research Unit, 25280 Solsona, Spain; yAgrotecnio, Universitat de Lleida, 25198 Lleida, Spain; zDepartment of HBJ, ETSEA, Universitat de Lleida, 25198 Lleida, Spain; aaCREAF, 08193 Cerdanyola del Vallès, Spain; bb Departament de Biologia Animal, de Biologia Vegetal i d’Ecologia, Universitat Autònoma Barcelona, 08193 Cerdanyola Del Vallès, Spain; ccInstituto Universitario de Investigación, CIBIO, University of Alicante, 03690 San Vicente del Raspeig, Spain; ddArea de Ecología, Facultad de Biología, Universidad de Murcia, 30100, Murcia, Spain; eeBritish Trust for Ornithology, Thetford, Norfolk IP24 2PU, United Kingdom; and ffCSIC, 08193 Cerdanyola del Vallès, Spain Edited by Claire Kremen, University of California, Berkeley, CA, and accepted by Editorial Board Member Ruth S. DeFries June 28, 2019 (received for review April 18, 2019) Agricultural landscape homogenization has detrimental effects on delivery of key ecosystem services to people (5). There is ample biodiversity and key ecosystem services. Increasing agricultural evidence that enhancing landscape heterogeneity by reversing landscape heterogeneity by increasing seminatural cover can help to mitigate biodiversity loss. However, the amount of seminatural Significance cover is generally low and difficult to increase in many intensively managed agricultural landscapes. We hypothesized that increasing Agricultural landscape homogenization is a major ongoing threat to the heterogeneity of the crop mosaic itself (hereafter “crop hetero- biodiversity and the delivery of key ecosystem services for human geneity”) can also have positive effects on biodiversity. In 8 con- well-being. It is well known that increasing the amount of semi- trasting regions of Europe and North America, we selected 435 natural cover in agricultural landscapes has a positive effect on landscapes along independent gradients of crop diversity and biodiversity. However, little is known about the role of the crop mean field size. Within each landscape, we selected 3 sampling sites mosaic itself. Crop heterogeneity in the landscape had a much in 1, 2, or 3 crop types. We sampled 7 taxa (plants, bees, butterflies, stronger effect on multitrophic diversity than the amount of semi- hoverflies, carabids, spiders, and birds) and calculated a synthetic natural cover in the landscape, across 435 agricultural landscapes index of multitrophic diversity at the landscape level. Increasing located in 8 European and North American regions. Increasing crop crop heterogeneity was more beneficial for multitrophic diversity heterogeneity can be an effective way to mitigate the impacts of than increasing seminatural cover. For instance, the effect of de- farming on biodiversity without taking land out of production. creasing mean field size from 5 to 2.8 ha was as strong as the effect of increasing seminatural cover from 0.5 to 11%. Decreasing mean Author contributions: G. Balent, B.P., F.B., T.T., V.B., G.S., A.O., L.B., J.-L.M., and L.F. de- field size benefited multitrophic diversity even in the absence of signed research; C.S., N.G., A.B.B., C.B., R.C., A.H., L.H., P.M., C. Vuillot, A.A., J.G., P.B., Y.C., seminatural vegetation between fields. Increasing the number C. Violle, D.G., G. Bota, I.B., G.L., B.G., A.V., F.C., A.G.-T., L.T., S.M., K.L., R.G., S.H., J.R., of crop types sampled had a positive effect on landscape-level X.O.S.-S., I.R., J. Bosch, J.A.B., A.R., M.Á.M.-G., J.M., R.M., A.G., and J. Baudry performed research; C.S., N.G., A.B.B., C.B., R.C., A.H., L.H., and P.M. analyzed data; and C.S. and N.G. multitrophic diversity. However, the effect of increasing crop di- wrote the paper. versity in the landscape surrounding fields sampled depended on The authors declare no conflict of interest. the amount of seminatural cover. Our study provides large-scale, This article is a PNAS Direct Submission. C.K. is a guest editor invited by the multitrophic, cross-regional evidence that increasing crop heteroge- Editorial Board. neity can be an effective way to increase biodiversity in agricultural Published under the PNAS license. landscapes without taking land out of agricultural production. A previous version of this work was published as part of Annika Hass’s 2018 Doctoral dissertation thesis, Farmland heterogeneity effects on biodiversity, community traits and | multitaxa biodiversity | crop mosaic | farmland | landscape insect pollination. Georg-August-Universität Göttingen, pp 16–67. complementation 1 To whom correspondence may be addressed. Email: clelia.sirami@inra.fr. 2 A.B.B., C.B., R.C., A.H., L.H., P.M., and C. Vuillot contributed equally to this work. A griculture dominates the world’s terrestrial area (1). Agri- cultural landscape homogenization through the decrease of seminatural cover, crop specialization, and field enlargement (2–4) 3 J.-L.M. and L.F. contributed equally to this work. This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. Downloaded by guest on April 8, 2021 1073/pnas.1906419116/-/DCSupplemental. represents a continuing worldwide threat to biodiversity and the Published online July 29, 2019. 16442–16447 | PNAS | August 13, 2019 | vol. 116 | no. 33 www.pnas.org/cgi/doi/10.1073/pnas.1906419116 the decline in seminatural cover can benefit biodiversity in ag- landscape. For instance, the “intermediate landscape-complexity” ricultural landscapes (6–8). However, the amount of seminatural hypothesis (17) predicts that the positive biodiversity-crop hetero- cover keeps decreasing in many agricultural landscapes, and the geneity relationship is stronger in landscapes with intermediate efficiency of policies focusing solely on maintaining or increasing amounts of seminatural cover (e.g., 5–20%) than in landscapes with seminatural cover has been questioned (9). little (e.g., <5%) or much seminatural cover (e.g., >20%; ref. 6). While half of the biodiversity in agricultural landscapes occurs Sampling over a wide range of landscapes may therefore be nec- exclusively in seminatural cover (10), the crop mosaic offers a wide essary to understand the general effect of crop heterogeneity on range of resources to the other half, including to species occurring farmland biodiversity. exclusively in crop fields and providing key ecosystem services, such The biodiversity–crop heterogeneity relationship may vary among as crop pollination or biological pest control (11, 12). It is therefore taxa (e.g., ref. 18). For instance, it may be more positive for species of increasing interest to evaluate whether enhancing landscape and taxa that have lower habitat area requirements (e.g., small spe- heterogeneity by increasing the heterogeneity of the crop mosaic cies; ref. 19) or higher habitat specialization levels (e.g., ref. 20). Al- itself (hereafter “crop heterogeneity”) can also benefit biodiversity though in-depth understanding of the effects of crop heterogeneity on (Fig. 1). There is growing pressure on agricultural land for food and each species or taxon is valuable, it is also critical to develop envi- energy production as well as for urbanization. Therefore, measures ronmental policies that are effective across a wide range of species to benefit biodiversity consisting of a rearrangement of the pro- (21). To achieve this, we here use a cross-regional sampling scheme in duction area, as opposed to measures focusing solely on its re- Europe and North America and a synthetic index integrating in- duction, could provide valuable new sustainable policy options. formation on multiple trophic groups to identify landscape patterns Crop heterogeneity can be decomposed into compositional het- that simultaneously increase the diversity of most taxa (22). erogeneity, i.e., the composition of the crop mosaic (e.g., crop di- We selected 435 landscapes along orthogonal gradients of versity), and configurational heterogeneity, i.e., the shape and spatial mean size and diversity of crop types available in the land- arrangement of fields (e.g., mean field size, ref. 13; see further ex- scape in 8 contrasting agricultural regions in France, the United planation in Methods). These 2 components of crop heterogeneity Kingdom, Germany, Spain, and Canada (Fig. S2.1, SI Appendix, may influence farmland biodiversity in several ways (see detailed section 2). In each landscape, we selected 3 sampling sites in 1, 2, alternative hypotheses in SI Appendix, section 1). First, increasing or 3 crop types. We sampled 7 taxa representing a wide range of ecological traits, functions, and trophic levels (plants, bees, butter- ECOLOGY crop diversity may benefit biodiversity if many species are specialists of distinct crop types (i.e., habitat specialization; Hyp-1a in SI Ap- flies, hoverflies, carabids, spiders, and birds) in each field. We then pendix, section 1; ref. 14). In that case, sampling increasing numbers computed a synthetic index of multitrophic diversity (Methods). We of crop types should lead to observing increasing levels of species tested the relative effects of mean field size, the number of crop diversity. Second, increasing crop diversity may also benefit bio- types sampled, the diversity of crop types available in the land- diversity through a landscape-level effect if many species require scape, and the amount of seminatural cover in the landscape on multiple resources provided by different crop types (i.e., landscape multitrophic diversity and on the species richness of taxonomic complementation; Hyp-1b in SI Appendix, section 1; ref. 15). In that groups. We also evaluated whether the effects of mean field size case, sampling a given number of crop types surrounded by in- and the diversity of crop types available in the landscape were creasing levels of crop diversity available in the landscape should nonlinear, nonadditive, and influenced by seminatural cover (see lead to observing increasing levels of species diversity. Third, detailed hypotheses in SI Appendix, section 1). decreasing mean field size may benefit biodiversity through a landscape-level effect if small fields provide easier access to adjacent Results and Discussion crop fields for many species (i.e., landscape complementation; Hyp-2a Our study provides large-scale evidence that crop heterogeneity in SI Appendix, section 1; refs. 15 and 16). In that case, sampling a is a major driver of multitrophic diversity in agricultural land- given number of fields surrounded by fields with decreasing mean scapes. The number of crop types sampled in the landscape, and sizes should lead to observing increasing levels of species diversity. the mean size and diversity of crop types available in the land- Biodiversity responses to crop heterogeneity may be nonlinear scape were consistently included in all models (Fig. 2A). To- and nonadditive. For instance, increasing the diversity of crops gether, they accounted for 61% of the explained variance in available in the landscape may benefit biodiversity in a given field multitrophic diversity, while seminatural cover accounted for only if fields are small enough for adjacent fields to be reached 24% (Fig. 2B). Interactions between seminatural cover and mean easily. Additionally, the effects of increasing crop heterogeneity on size/crop diversity of fields available in the landscape also biodiversity may depend on the amount of seminatural cover in the accounted for an important part of the explained variance (15%), indicating that the effects of crop heterogeneity were modulated by the amount of seminatural cover in the landscape (Fig. 3). The effects of crop heterogeneity on multitrophic di- Tradional representaon Alternave representaon A B versity were consistent across the 8 European and North American regions (Fig. 4). The effects of crop heterogeneity on Decreasing mean field size vegetation between fields the species richness of taxonomic groups were similar to their Increasing semi-natural effects on multitrophic diversity and similar across the 7 taxa (Fig. 5 and Fig. S5.2, SI Appendix, section 5). They hold true when considering either landscape-level or field-level multi- trophic diversity, including when focusing only on cereal fields, the most dominant crop type across our 8 regions (Table S5.11, SI Appendix, section 5). Their effects were also unchanged when potential confounding factors such as the identity of crop types sampled, land-use intensity within fields sampled (i.e., an index combining data on plowing, fertilizer, herbicide, and insecticide), Increasing semi-natural cover Increasing crop diversity the composition of the crop mosaic, grassland cover, or hedge- Semi-natural cover row length available in the landscape were taken into account in Semi-natural vegetaon between fields Crop types our analyses (SI Appendix, section 5). Fig. 1. (A) Traditional and (B) alternative representations of agricultural landscape heterogeneity, focusing either on seminatural heterogeneity or Consistent Positive Effects of Decreasing Mean Field Size on Multitrophic crop heterogeneity, are associated with distinct hypotheses (see details in SI Diversity. Decreasing mean field size was the main driver of mul- titrophic diversity variations, mean field size and mean field size2 Downloaded by guest on April 8, 2021 Appendix, section 1; large squares represent landscapes). Adapted with permission from ref. 13. together accounting for 47.4% of the explained variance in Sirami et al. PNAS | August 13, 2019 | vol. 116 | no. 33 | 16443 A 1.00 B 100 Fig. 2. (A) Importance and (B) relative effect of the CD, the number of crop types sampled (Crop Nb), model averaging approach Relative importance in the SNC *** Variance explained in the 0.80 80 MFS, SNC, and interaction terms (CD:SNC, MFS:SNC) averaged model (%) MFS:SNC on multitrophic diversity. P < 0.1; *P < 0.05; **P < 0.60 CD:SNC * 0.01; ***P < 0.001. Variables are grouped in 3 com- 60 Crop Nb ** ponents: orange = crop heterogeneity (MFS, MFS2, 0.40 CD CD, CD2, MFS:CD, Crop Nb), green = SNC (SNC, SNC2), 40 MFS² ° blue = interactive effects between crop heteroge- 0.20 neity and SNC (CD:SNC, MFS:SNC, CD:MFS:SNC). The 20 relative importance of variables is estimated as the MFS * proportion of submodels where the variable was 0.00 selected. The variance explained by each variable MFS² MFS SNC² Crop Nb CD² CD MFS :SNC SNC CD:MFS CD :SNC CD:MFS:SNC 0 corresponds to the ratio between its parameter es- -15 -10 -5 0 5 10 timate and the sum of all parameter estimates based Parameter esmates on a model averaging approach applied to model 1. multitrophic diversity (Fig. 2B). The effect of decreasing mean surrounding sampled fields were consistently included in all models field size from 5 to 2.75 ha was as strong as the effect of increasing (Fig. 2A). This result suggests that both field-level (i.e., habitat seminatural cover from 0.5 to 11% of the landscape (Fig. 3B). specialization) and landscape-level processes (i.e., landscape com- Such a positive effect of decreasing mean field size on multi- plementation and/or spill-over) can contribute to the effect of crop trophic diversity is consistent with the hypothesis that smaller diversity on multitrophic diversity (SI Appendix, sections 1 and 4). fields provide easier access to multiple cover patches, in particular Increasing the number of crop types sampled had a significant for species that require resources occurring in different cover types positive effect accounting for 13% of the explained variance in (landscape complementation; refs. 15 and 16). The positive effect landscape-level multitrophic diversity (Fig. 2B). This result of decreasing mean field size was particularly clear and strong confirms that increasing crop diversity results in a larger number when mean field size fell below 6 ha (93% of landscapes studied). of distinct habitats, and therefore higher biodiversity levels by Although the strength of this effect varied significantly among increasing the number of specialist species in the landscape regions, decreasing mean field size had a consistent positive ef- (Hyp-1a in SI Appendix, section 1; ref. 20). fect across all regions studied (Fig. 4 and SI Appendix, section 5). The main effect of increasing the diversity of crop types avail- It was also consistently positive across all taxa considered sepa- able in the landscape was nonsignificant but significantly mediated rately, from primary producers to predators (Fig. 5 and SI Ap- pendix, section 5). Previous studies have already reported by seminatural cover. These effects were similar across all regions positive effects of decreasing mean field size on the diversity of (Fig. 4). Together, the diversity of crop types available in the several taxa considered separately (e.g., refs. 23–27). Our study, landscape and its interaction with seminatural cover accounted based on multiple regions and multiple trophic groups, shows that for 10% of the explained variance in multitrophic diversity the benefits of decreasing mean field size can be generalized to (Fig. 2B). The landscape-level effect of increasing crop diversity on multitrophic diversity across a wide range of agricultural regions. multitrophic diversity ranged from negative in landscapes with low Previous studies suggested that the positive effect of de- seminatural cover to positive in landscapes with high seminatural creasing mean field size on multitrophic diversity may be pri- cover (Fig. 3A). This result is consistent with the variability of effects marily due to the presence of seminatural vegetation between observed across previous studies (e.g., refs. 23, 29, and 30). To fields (e.g., refs. 23, 25, and 26). To test this hypothesis, we se- further explore this interaction, we used the same moving window lected a subset of landscapes for which mean field size and the modeling approach described above (SI Appendix, section 5). This length of seminatural vegetation between fields were un- analysis confirmed that the landscape-level effect of increasing crop correlated (SI Appendix, section 5). The analysis, based on 274 diversity on multitrophic diversity was positive in landscapes with landscapes, showed that the positive effect of increasing mean more than 11% seminatural cover (i.e., 50% of landscapes included field size on multitrophic diversity cannot be explained solely by in our study), nonsignificant in landscapes with 4–11% seminatural the increase in the length of seminatural vegetation between cover (i.e., 34% of landscapes), and negative in landscapes with less fields. Increasing the length of seminatural vegetation between than 4% seminatural cover (i.e., 16% of landscapes; Fig. S5.5.A, fields had a positive effect on multitrophic diversity but including SI Appendix, section 5). this effect in our model did not change the effect of mean field size on multitrophic diversity (Table S5.8, SI Appendix, section 5). This result suggests that smaller fields benefit multitrophic diversity even in the absence of seminatural vegetation between fields. A B Finally, the presence of the interaction term between mean field size and seminatural cover in our model (Fig. 2A) suggests Multrophic diversity that the effect of mean field size on multitrophic diversity tends to be modulated by the amount of seminatural cover available in the landscape (Fig. 3B). To further explore this interaction, we used a moving window modeling approach (ref. 28; SI Appendix, section 5). This analysis confirmed that decreasing mean field size had a consistent positive effect on multitrophic diversity along the gradient of seminatural cover. Moreover, it suggested that this effect is stronger when seminatural cover is below 8%, i.e., when seminatural cover is too scarce to provide access to the multiple resources required by most species occurring in agri- cultural landscapes (Fig. S5.5.B, SI Appendix, section 5). Fig. 3. Interactive effects of (A) CD and SNC and (B) MFS and SNC on multitrophic diversity. The color gradient corresponds to multitrophic di- Complex Effects of Increasing Crop Diversity on Multitrophic versity values, ranging from low values (blue) to high values (red). Slopes are Downloaded by guest on April 8, 2021 Diversity. The number of crop types sampled in each landscape based on the outputs of model 1 (see Methods). Gray dots show the overall and the diversity of crop types available in the landscape gradients in CD, MFS, and SNC across the 435 landscapes located in 8 regions. 16444 | www.pnas.org/cgi/doi/10.1073/pnas.1906419116 Sirami et al. Armorique Camargue agricultural landscapes. Policy measures favoring crop heteroge- Multitrophic diversity 20 40 60 80 80 Coteaux East Anglia neity may be more easily implemented than policies to increase Goettingen seminatural cover or reduce chemical use (40). Associated with 60 Lleida Ontario PVDS adequate economic incentives, they may also be more favorably 40 perceived by farmers and thus lead to higher uptake than mea- 20 sures requiring farmers to take land out of production (39). Such measures may also contribute to the development of frameworks 0 0 0.0 0.5 1.0 1.5 2.0 0 2 4 6 8 10 12 that reward farmers for sustainable land stewardship. CD MFS (ha) We observed a consistent effect of crop heterogeneity on species diversity across 7 taxa representing a wide range of ecological traits, Multitrophic diversity 20 40 60 80 80 functions, and trophic levels (plants, bees, butterflies, hoverflies, 60 carabids, spiders, and birds; Fig. 5). We observed landscapes where 6 or even all 7 taxa reached the threshold of 60% of the maximum 40 species richness observed within a given region (Fig. 4). Our study 20 therefore suggests that policies to increase crop heterogeneity 0 0 0 10 20 30 40 50 0 10 20 30 40 50 60 would be an effective way to increase the diversity of all compo- SNC (%) CD:SNC nents of biodiversity simultaneously and restore multitrophic bio- diversity in agricultural landscapes. Fig. 4. Effects of CD, MFS, SNC, and the interaction between CD and SNC (CD:SNC) on multitrophic diversity in different regions. Slopes are based on Finally, our results can contribute to the development of policies the outputs of model 2 (see Methods) including a random effect of region adapted to different landscape contexts. For instance, our results on these 4 slopes (n = 435 landscapes). Colors indicate the region. suggest that policy measures aimed at decreasing field sizes to be- low 6 ha may be particularly effective to promote multitrophic di- versity in agricultural landscapes, especially in landscapes where The positive landscape-level effect of increasing crop diversity on seminatural cover is below 8%. Our results also caution against a ECOLOGY multitrophic diversity observed in landscapes with more than 11% “blind” increase of crop diversity. Measures aimed at increasing seminatural cover supports the “landscape complementation” hy- crop diversity may be effective to promote multitrophic diversity in pothesis (Hyp-1b in SI Appendix, section 1). This finding is con- landscapes where seminatural cover exceeds 11%. However, they sistent with the fact that a diverse crop matrix provides a temporal are more likely to be effective in promoting multitrophic diversity continuity of food sources (31) while seminatural patches provide across all agricultural landscapes if combined with measures pro- stable resources, for example, for nesting or shelter (e.g., ref. 32). moting the restoration or maintenance of seminatural cover. Such complementation among multiple cover types has been de- scribed for several species (e.g., refs. 33 and 34). Our study, based Conclusion on multiple regions and multiple trophic groups, shows that the positive landscape-level effect of increasing crop diversity can be Our study demonstrates the importance of crop heterogeneity generalized to multitrophic diversity across many agricultural for multitrophic diversity in agricultural landscapes: The effect landscapes (50% of landscapes included in our study). of maintaining/increasing crop heterogeneity is likely to be as The negative landscape-level effect of increasing crop diversity important as the effect of maintaining/increasing seminatural on multitrophic diversity in landscapes with less than 4% semi- cover. This finding suggests that field enlargement and crop natural cover supports the “minimum total habitat area re- specialization, especially the former, have been underestimated quirement” hypothesis (Hyp-1c in SI Appendix, section 1). This drivers of past and ongoing biodiversity declines. More impor- finding is consistent with the fact that landscape simplification tantly, our study shows that increasing crop heterogeneity rep- tends to filter out species with large body sizes (35), which also resents a major potential lever to increase synergies between have high minimum total habitat area requirements (36), and food production and biodiversity conservation. may therefore require high amount of a single crop type. How- ever, taxa included in the present study were associated with a wide range of ecological traits, and therefore a wide range of minimum total habitat area requirements. That they showed a 0.6 0.6 bee bird consistent response to crop diversity and the interaction of crop Species richness butterfly carabid −0.6 −0.2 0.2 0.2 diversity and seminatural cover (Fig. 5) suggests that the minimum plant spider total habitat area requirement hypothesis is unlikely to be the sole syrphid −0.2 mechanism contributing to our results. Future research is needed to identify additional mechanisms and conditions under which in- −0.6 creasing crop diversity leads to a consistent net positive effect on multitrophic diversity, i.e., a positive effect of habitat specialization 0.0 0.5 1.0 1.5 2.0 0 2 4 6 8 10 12 CD MFS (ha) plus landscape complementation processes. 0.6 0.6 Implications for Agricultural Policies. Our study has important im- Species richness −0.6 −0.2 0.2 0.2 plications for large-scale policy schemes implemented across a wide range of contexts such as the European Common Agricultural −0.2 Policy and its recent greening (21), the Canadian Agriculture Policy Frameworks (37), or the United States Farm Bill (38). −0.6 First, our results suggest that increasing crop heterogeneity may have a similar or greater benefit for multitrophic diversity to 0 10 20 30 40 50 0 10 20 30 40 50 60 SNC (%) CD:SNC increasing seminatural cover (Fig. 2B) or even decreasing field- level land-use intensity (ref. 16; Table S5.12, SI Appendix, section Fig. 5. Effects of CD, MFS, SNC, and the interaction between CD and SNC 5). Given current challenges to increase seminatural cover and (CD:SNC) on the landscape-level species richness of taxonomic groups. Slopes limit chemical use in agricultural landscapes (39), policies aiming are based on the outputs of model 10 (see SI Appendix, section 5) including a at increasing crop heterogeneity may represent an effective and Downloaded by guest on April 8, 2021 random effect of taxon on these 4 slopes (n = 435 landscapes). Colors indicate complementary way to improve biodiversity conservation in the taxon. Sirami et al. PNAS | August 13, 2019 | vol. 116 | no. 33 | 16445 Methods the given threshold to produce a multidiversity index (T60.landscape) for Region, Landscape, and Sampling Site Selection. We selected 8 agricultural each landscape, based on the following formula: regions (Armorique, Camargue, Coteaux de Gascogne, and Plaine et Val de Sèvre in France, East Anglia in the United Kingdom, Goettingen in Germany, Lleida in 1 X i=n MultidiversityðTx. landscapeÞ = ½SR i > ðx × SR max. region jÞ, Spain, and Eastern Ontario in Canada; Fig. S2.1, SI Appendix, section 2) be- n i=1 longing to 6 different ecoregions (41) and differing in topography, climate, field shapes, and agricultural cover types and products (e.g., rice, dairy, tree crops). where n is the number of taxa for which data were available in a given We used the best spatial data available within each region before field landscape (SI Appendix, section 3), SRi is the number of species for taxon i, x work to identify all 1 km × 1-km rural landscapes, i.e., those dominated by is the minimum threshold to be reached, and SRmax.region j is the maximum species richness for taxon i in the region the landscape considered belonging agricultural cover (>60%, including all crops and grassland managed for to. This multidiversity index ranges between 0 and 1 (SI Appendix, section 3). agricultural production). We then developed a protocol to select a combi- For simplicity, we hereafter refer to “landscape-level multitrophic diversity” nation of landscapes that maximized the gradients of crop diversity and rather than T60.landscape. mean field size, while minimizing the correlation between them (42). Crop diversity may theoretically be constrained by the number and size of fields in landscapes with large fields. However, in our dataset, mean field size was Mean Field Size and Diversity of Crop Types in the Landscape. We used a smaller than 12 ha and was therefore not a limiting factor for crop diversity standardized protocol across all regions to produce land-cover maps allowing within the 1 km × 1-km landscapes. We selected between 32 and 93 land- us to compare consistency of effects across the 8 regions (SI Appendix, section scapes within each region, totaling 435 landscapes across all regions. 4). We conducted extensive ground-truthing surveys during the field seasons to map all fields, linear elements between adjacent fields, and non- We selected 3 sampling sites within each landscape, totaling 1,305 sam- agricultural covers. We built a common land-cover classification for the 8 pling sites across all regions. The number of crop types sampled ranged from 1 regions (SI Appendix, section 4). We then used these standardized, detailed to 3 per landscape. Where feasible, we located sampling sites in dominant maps to calculate 4 explanatory variables for each landscape: diversity of agricultural cover types within each region (e.g., wheat fields and oilseed crop types in the landscapes, mean field size, seminatural cover, and total rape in Goettingen). When this was not feasible, we located sampling sites length of seminatural linear elements between fields. in agricultural cover types that were accessible within a given landscape (SI Appendix, section 3). The 3 sampling sites were at least 200 m from each We calculated the Shannon diversity of agricultural cover types (hereafter other, at least 50 m from the border of the landscape, and at least 50 m “the diversity of crop types in the landscape;” CD) and the mean field size in from patches of nonagricultural cover types such as forests and urban hectares (MFS). Neither CD nor MFS was correlated with local land-use in- areas. tensity (an index combining data on plowing, fertilizer, herbicide, and in- secticide) or the overall composition of the crop mosaic across all regions (SI Appendix, section 5). CD and MFS were moderately correlated with the type Multitaxa Sampling. We selected 7 taxa representing a wide range of eco- of crops sampled in some regions and MFS was moderately correlated with logical traits, functions, and trophic levels which, combined into a multi- the proportion of grassland in the crop mosaic, but none of these correla- diversity index (see below), represent a proxy for multitrophic diversity: tions affected our conclusions (SI Appendix, section 5). We calculated the plants, bees, butterflies, hoverflies, carabids, spiders, and birds. All taxa were percentage of seminatural cover types, i.e., woodland, open land, and sampled using standardized sampling protocols across all regions, allowing us wetland (SNC), in each landscape. We also calculated the total length of to test the consistency of effects across the 8 regions (SI Appendix, section 3). linear seminatural elements between fields, e.g., hedgerows, grassy margins At each sampling site, we selected 2 parallel 50-m “transects,” one located (SNL; measured in meters). SNL and MFS were highly correlated in some at the field edge and the other inside the field 25 m away from the first regions (Table S5.6, SI Appendix, section 5). As a result, we did not include transect (Figs. S3.1 and S3.2, SI Appendix, section 3). Birds were sampled SNL in the main analyses and only tested the relative effect of MFS and SNL using point counts centered on the field-edge transect. Plants were surveyed using a subset of our dataset for which MFS and SNL were not strongly along both transects. Butterflies were surveyed visually using timed walks correlated (SI Appendix, section 5). along both transects. Bees and hoverflies were sampled using colored pan traps on poles erected at each end and in the center of all transects. Carabids Data Analysis. We first tested the effect of crop heterogeneity on multitrophic and spiders were sampled using pitfall traps installed at each end of all diversity (model 1). We fitted a linear mixed model with restricted maximum transects. Captured arthropods were preserved in ethanol prior to identifi- likelihood using the landscape-level multidiversity index (T60.landscape) as cation. Multiple survey visits were conducted during the season when rele- the response variable. We included the number of crop types sampled per vant (SI Appendix, section 3). Each landscape was sampled during 1 y and landscape (CropNb), CD, MFS, and SNC as explanatory variables (see alter- sampling of landscapes was distributed across 2 y within each region, be- native hypotheses in SI Appendix, section 1). We included both interaction tween 2011 and 2014 (Table S3.1, SI Appendix, section 3). effects and quadratic effects. Due to a positive skew in the distribution of We identified more than 167,000 individuals from 2,795 species (Table MFS, we used log MFS in all analyses. To reflect the large-scale spatial and S3.2, SI Appendix, section 3). For each taxon, we calculated species richness temporal structure of our dataset, we added sampling year (Year), nested at the landscape level, i.e., across all 3 sampling sites and across all visits within study region (Region), as a random effect. To reflect the spatial when multiple survey visits were conducted. The average species richness structure of our dataset within each region, we included the longitude and per landscape varied greatly among taxa, from 5.4 for butterflies to 44.9 latitude of the center of each landscape (Lat, Lon) as covariates. We stan- for plants. Correlations in average species richness between pairs of taxa dardized all fixed effects to allow for a direct comparison of estimates. were weak (<0.41), with an average correlation of 0.07 (Table S3.3, SI Model 1: lmer (T60.landscape ∼ CD * MFS * SNC + CD2 + MFS2 + SNC2 + CropNb + Appendix, section 3). Lat + Lon + (1jRegion/Year)). Then, we added random effects for region on the slopes of the diversity of crop types in the landscape, MFS, SNC, as well as the Multitrophic Diversity Index. A classical approach in the literature consists of interaction between the CD and SNC (model 2). We assumed that the effects calculating the average, standardized diversity across taxa (43). However, this of region on the intercept and slopes were uncorrelated. To test whether approach has limitations (SI Appendix, section 3). Although very high/low region had a significant effect on the slope of either the CD, MFS, or SNC, as values imply that all taxa exhibit high/low diversity, intermediate values are well as the interaction between crop diversity and SNC, we used the function difficult to interpret as they may correspond to situations where 1) diversity exactRLRT from package RLRsim. values are intermediate for all taxa, or 2) diversity values are high for some Model 2: lmer (T60.landscape ∼ CD * MFS * SNC + CD2 + MFS2 + SNC2 + CropNb + taxa and low for others, i.e., trade-offs among taxa. To overcome this limi- Lat + Lon + (1jRegion/Year) + (0+CDjRegion)) + (0+MFSjRegion) + (0+SNCjRegion) + tation, we used a threshold approach initially developed to aggregate (0+CD:SNCjRegion)). We then tested the effects of crop heterogeneity on the SR multiple ecosystem functions (22). of taxonomic groups (model 3). To do this, we fitted a similar model, using the For each taxon and each region, we identified the maximum species landscape-level SR of taxonomic groups standardized within each taxon and richness observed across all landscapes. We used the 95th percentile as the region as the response variable. To reflect that species pools vary between taxa, maximum observed species richness (hereafter “SR max”) to minimize the we added taxon as a random effect. effect of outliers. Next, we identified which landscapes attained a given Model 3: lmer (SR ∼ CD*MFS*SNC + CD2 + MFS2 + SNC2 + CropNb + Lat + Lon + threshold (x) of SR max. We chose to use an intermediate threshold of 60% (1jRegion/Year) + (1jTaxon)). Then we added random effects for taxon on the because intermediate thresholds have been shown to provide an effective slopes of CD, MFS, SNC, as well as the interaction between CD and SNC (model 4). Downloaded by guest on April 8, 2021 measure of multitrophic diversity in agricultural landscapes (ref. 43 and SI We assumed that the effects of taxon on the intercept and slopes were un- Appendix, section 3). We then tallied the proportion of taxa that exceeded correlated. To test whether taxon had a significant effect on the slope of either 16446 | www.pnas.org/cgi/doi/10.1073/pnas.1906419116 Sirami et al. CD, MFS, SNC or the interaction between CD and SNC, we used the function Data Accessibility. All protocols, datasets, and R codes used in this article can exactRLRT from package RLRsim. be made available upon request to the corresponding author. Model 4: lmer (SR ∼ CD * MFS * SNC + CD2 + MFS2 + SNC2 + CropNb + Lat + Lon + (1jTaxon) + (1jRegion/Year) + (0+CDj Taxon)) + (0+MFSjTaxon) + (0+SNCj Taxon) + ACKNOWLEDGMENTS. This research was funded by the ERA-Net BiodivERsA, (0+CD:SNCj Taxon)). We fitted all models with the R lme4 package using with the national funders French National Research Agency (ANR-11-EBID- LMER (44), we removed outliers using function romr.fnc from package 0004), German Ministry of Research and Education, German Research LMERConvenienceFunctions (45), and we ran diagnostic tools to verify that Foundation and Spanish Ministry of Economy and Competitiveness, part of residuals were independently and normally distributed, and showed no spatial the 2011 BiodivERsA call for research proposals. The UK component of this autocorrelation. For each model, a multimodel inference procedure was ap- research was funded by the UK Government Department of the Environment, plied using the R MuMIn package (46). This method allowed us to perform Food and Rural Affairs (Defra), as Project WC1034. The Canadian component of this research was funded by a Natural Sciences and Engineering Research model selection by creating a set of models with all possible combinations of Council of Canada Strategic Project, the Canada Foundation for Innovation, the initial variables and sorting them according to the Akaike Information Environment Canada, and Agriculture and Agri-Food Canada. N.G. was Criterion (AIC) fitted with Maximum Likelihood (47). We selected all models supported by the AgreenSkills+ Fellowship programme which has received with ΔAIC <2 and used the model averaging approach using LMER to estimate funding from the EU’s Seventh Framework Programme under Grant Agree- parameters and associated P values, using the function model.avg. We ran all ment FP7-609398 (AgreenSkills+ contract). A.G.-T. (Juan de la Cierva Fellow, analyses using the software R 3.4.0 (48). JCI-2012-12089) was funded by Ministerio de Economía y Competitividad We ran additional analyses to check that the composition of the crop (Spain). C. Violle was supported by the European Research Council Starting mosaic, the proportion of grassland in the crop mosaic, and the amount of Grant Project “Ecophysiological and biophysical constraints on domestication seminatural vegetation occurring between fields did not affect our conclu- of crop plants” (Grant ERC-StG-2014-639706-CONSTRAINTS). A.R.’s position at sions (SI Appendix, section 5). We also ran complementary analyses using the University of Alicante is funded by the “Vicerrectorado de Investigación y field-level multidiversity (T60.field) as the response variable––instead of the Transferencia de Conocimiento.” We thank the hundreds of farmers and farm owners from all 8 regions who graciously permitted us to work on their lands. In landscape-level multidiversity index (T60.landscape)––to check that our re- addition to the coauthors, the project involved direct assistance from more than sults hold true at the field level, in particular within a subset of cereal fields, 150 individuals for geomatics analyses, field sampling, and species identifica- and that the type of crop sampled or the level of land-use intensity within tion. We thank all of them for their huge contribution to the FarmLand project sampled fields did not affect our conclusions (SI Appendix, section 5). Finally, and this study. We are grateful to the GLEL (Geomatics and Landscape Ecology we used a moving window analysis to identify potential discontinuities in Laboratory) Friday Discussion Group for very helpful input. Finally, we would multitrophic diversity response to CD and MFS along the gradient of SNC (SI like to thank the editor, Sarina Macfadyen, and another anonymous reviewer ECOLOGY Appendix, section 5). for their very constructive comments on an earlier version of this paper. 1. D. Tilman et al., Forecasting agriculturally driven global environmental change. Sci- 26. M.  Sálek et al., Bringing diversity back to agriculture: Smaller fields and non-crop ence 292, 281–284 (2001). elements enhance biodiversity in intensively managed arable farmlands. Ecol. Indic. 2. E. V. White, D. P. Roy, A contemporary decennial examination of changing agricul- 90, 65–73 (2018). tural field sizes using Landsat time series data. Geo 2, 33–54 (2015). 27. A. L. Hass et al., Landscape configurational heterogeneity by small-scale agriculture, 3. C. J. Barr, M. K. Gillespie, Estimating hedgerow length and pattern characteristics in not crop diversity, maintains pollinators and plant reproduction in western Europe. Great Britain using Countryside Survey data. J. Environ. Manage. 60, 23–32 (2000). Proc. R. Soc. B 285, 20172242 (2018). 4. J. Aguilar et al., Crop species diversity changes in the United States: 1978-2012. PLoS 28. M. Berdugo et al., Aridity preferences alter the relative importance of abiotic and biotic One 10, e0136580 (2015). drivers on plant species abundance in global drylands. J. Ecol. 107, 190–202 (2019). 5. T. Newbold et al., Global effects of land use on local terrestrial biodiversity. Nature 29. J. Josefsson, Å. Berg, M. Hiron, T. Pärt, S. Eggers, Sensitivity of the farmland bird community 520, 45–50 (2015). to crop diversification in Sweden: Does the CAP fit? J. Appl. Ecol. 54, 518–526 (2017). 6. P. Batáry, J. Fischer, A. Báldi, T. O. Crist, T. Tscharntke, Does habitat heterogeneity 30. E. M. Olimpi, S. M. Philpott, Agroecological farming practices promote bats. Agric. increase farmland biodiversity? Front. Ecol. Environ. 9, 152–153 (2011). Ecosyst. Environ. 265, 282–291 (2018). 7. T. Miyashita, Y. Chishiki, S. R. Takagi, Landscape heterogeneity at multiple spatial 31. N. A. Schellhorn, V. Gagic, R. Bommarco, Time will tell: Resource continuity bolsters scales enhances spider species richness in an agricultural landscape. Popul. Ecol. 54, ecosystem services. Trends Ecol. Evol. (Amst.) 30, 524–530 (2015). 573–581 (2012). 32. C. Sirami, L. Brotons, J. Martin, Woodlarks Lullula arborea and landscape heteroge- 8. D. Perovic et al., Configurational landscape heterogeneity shapes functional com- neity created by land abandonment. Bird Study 58, 99–106 (2011). munity composition of grassland butterflies. J. Appl. Ecol. 52, 505–513 (2015). 33. S. E. Pope, L. Fahrig, N. G. Merriam, Landscape complementation and metapopulation 9. P. Batáry, L. V. Dicks, D. Kleijn, W. J. Sutherland, The role of agri-environment schemes effects on leopard frog populations. Ecology 81, 2498–2508 (2000). in conservation and environmental management. Conserv. Biol. 29, 1006–1016 (2015). 34. T. Mueller, N. Selva, E. Pugacewicz, E. Prins, Scale-sensitive landscape complementation 10. G. Lüscher et al., Farmland biodiversity and agricultural management on 237 farms in determines habitat suitability for a territorial generalist. Ecography 32, 345–353 (2009). 13 European and two African regions. Ecology 97, 1625 (2016). 35. S. Gámez-Virués et al., Landscape simplification filters species traits and drives biotic 11. A. Holzschuh, C. F. Dormann, T. Tscharntke, I. Steffan-Dewenter, Mass-flowering homogenization. Nat. Commun. 6, 8568 (2015). crops enhance wild bee abundance. Oecologia 172, 477–484 (2013). 36. M. Baguette, V. Stevens, Predicting minimum area requirements of butterflies using 12. L. Raymond et al., Immature hoverflies overwinter in cultivated fields and may signifi- life-history traits. J. Insect. Conserv. 17, 645–652 (2013). cantly control aphid populations in autumn. Agric. Ecosyst. Environ. 185, 99–105 (2014). 37. B. J. Deaton, P. Boxall, Canadian agricultural policy in the twenty‐first century: Looking 13. L. Fahrig et al., Functional landscape heterogeneity and animal biodiversity in agri- back and going forward. Can. J. Agric. Econ. Can. Daposagroeconomie 65, 519–522 (2017). cultural landscapes. Ecol. Lett. 14, 101–112 (2011). 38. A. Reimer, Ecological modernization in U.S. agri-environmental programs: Trends in 14. A.-C. Weibull, Ö. Östman, Å. Granqvist, Species richness in agroecosystems: The effect of the 2014 farm bill. Land Use Policy 47, 209–217 (2015). landscape, habitat and farm management. Biodivers. Conserv. 12, 1335–1355 (2003). 39. G. Pe’er et al., Adding some green to the greening: Improving the EU’s ecological 15. J. B. Dunning, B. J. Danielson, H. R. Pulliam, Ecological processes that affect pop- focus areas for biodiversity and farmers. Conserv. Lett. 10, 517–530 (2017). ulations in complex landscapes. Oikos 65, 169–175 (1992). 40. C. Rodríguez, K. Wiegand, Evaluating the trade-off between machinery efficiency 16. P. Batáry et al., The former Iron Curtain still drives biodiversity-profit trade-offs in and loss of biodiversity-friendly habitats in arable landscapes: The role of field size. German agriculture. Nat. Ecol. Evol. 1, 1279–1284 (2017). Agric. Ecosyst. Environ. 129, 361–366 (2009). 17. T. Tscharntke et al., Landscape moderation of biodiversity patterns and processes– 41. D. M. Olson et al., Terrestrial ecoregions of the world: A new map of life on Earth a eight hypotheses. Biol. Rev. Camb. Philos. Soc. 87, 661–685 (2012). new global map of terrestrial ecoregions provides an innovative tool for conserving 18. F. Burel, A. Butet, Y. R. Delettre, N. M. de la Pena, Differential response of selected taxa to biodiversity. Bioscience 51, 933–938 (2001). landscape context and agricultural intensification. Landsc. Urban Plan. 67, 195–204 (2004). 42. J. Pasher et al., Optimizing landscape selection for estimating relative effects of 19. A. Ponjoan, G. Bota, S. Mañosa, Ranging behaviour of little bustard males, Tetrax landscape variables on ecological responses. Landsc. Ecol. 28, 371–383 (2013). tetrax, in the lekking grounds. Behav. Processes 91, 35–40 (2012). 43. J. E. K. Byrnes et al., Investigating the relationship between biodiversity and ecosystem 20. S. Gaba, B. Chauvel, F. Dessaint, V. Bretagnolle, S. Petit, Weed species richness in winter multifunctionality: Challenges and solutions. Methods Ecol. Evol. 5, 111–124 (2014). wheat increases with landscape heterogeneity. Agric. Ecosyst. Environ. 138, 318–323 (2010). 44. D. Bates, M. Mächler, B. M. Bolker, S. C. Walker, Fitting linear mixed-effects models 21. G. Pe’er et al., Agriculture policy. EU agricultural reform fails on biodiversity. Science using lme4. J. Stat. Softw. 67, 1–48 (2015). 344, 1090–1092 (2014). 45. A. Tremblay, J. Ransijn, LMERConvenienceFunctions: Model selection and post-hoc 22. E. Allan et al., Interannual variation in land-use intensity enhances grassland multi- analysis for (G) LMER models, R package Version 2.10 (2015). https://CRAN.R-project.org/ diversity. Proc. Natl. Acad. Sci. U.S.A. 111, 308–313 (2014). package=LMERConvenienceFunctions. Accessed 14 April 2019. 23. L. Fahrig et al., Farmlands with smaller crop fields have higher within-field bio- 46. K. Barton, MuMIn: Multi-Model Inference, R Package Version 0.12.0. Httpr-Forg-Proj diversity. Agric. Ecosyst. Environ. 200, 219–234 (2015). (2009). https://ci.nii.ac.jp/naid/10030574914/. Accessed 1 August 2018. 24. S. J. Collins, L. Fahrig, Responses of anurans to composition and configuration of 47. A. Zuur, E. N. Ieno, N. Walker, A. A. Saveliev, G. M. Smith, Mixed Effects Models and agricultural landscapes. Agric. Ecosyst. Environ. 239, 399–409 (2017). Extensions in Ecology with R (Springer, New York, 2009). 25. L. Monck-Whipp, A. E. Martin, C. M. Francis, L. Fahrig, Farmland heterogeneity 48. R Core Team, R: A Language and Environment for Statistical Computing (R Founda- Downloaded by guest on April 8, 2021 benefits bats in agricultural landscapes. Agric. Ecosyst. Environ. 253, 131–139 (2018). tion for Statistical Computing, Vienna, 2016). Sirami et al. PNAS | August 13, 2019 | vol. 116 | no. 33 | 16447