Diversity increases yield but reduces harvest index in crop mixtures

Abstract

Resource allocation to reproduction is a critical trait for plant fitness^1,2. This trait, called harvest index in the agricultural context^3–5, determines how plant biomass is converted to seed yield and consequently financial revenue of numerous major staple crops. While plant diversity has been demonstrated to increase plant biomass^6–8, plant diversity effects on seed yield of crops are ambiguous⁹ and dependent on the production syndrome¹⁰. This discrepancy might be explained through changes in the proportion of resources invested into reproduction in response to changes in plant diversity, namely through changes of species interactions and microenvironmental conditions^11–14. Here we show that increasing crop plant diversity from monocultures over 2- to 4-species mixtures increased annual primary productivity, resulting in overall higher plant biomass and, to a lesser extent, higher seed yield in mixtures compared with monocultures. The difference between the two responses to diversity was due to a reduced harvest index of the eight tested crop species in mixtures, possibly because their common cultivars have been bred for maximum performance in monoculture. While crop diversification provides a sustainable measure of agricultural intensification¹⁵, the use of currently available cultivars may compromise larger gains in seed yield. We therefore advocate regional breeding programs for crop varieties to be used in mixtures that should exploit complementarity¹⁶ among crop species.

Based on the vast ecological literature demonstrating positive relationships between plant diversity and annual primary productivity^17,18, increasing crop plant diversity through intercropping, i.e. the simultaneous cultivation of more than one crop species on the same land, has been proposed as a promising sustainable intensification measure in agriculture^15,19,20. However, although evidence on positive crop plant diversity–seed yield relationships is strong, the effect sizes are heterogenous and depend, among others, on the temporal overlap of the different species in the field^9,10,21. This could be due to non-linear reproductive allocation patterns, where increased annual primary productivity in mixtures would not translate into corresponding increases in seed yield.

The amount of resources allocated to seeds is a critical component of plant fitness^1,2,22–24 and directly determines grain yield and the economic value of annual grain crops, including the major staple crops wheat, maize, rice, soybeans, beans and barley^24–26. For crops, resource allocation has therefore been a target trait under selection during plant domestication²⁷ and modern plant breeding³. In general, reproductive allocation is allometric²⁸, i.e. seed yield increases alongside vegetative plant biomass^29,30. However, varying abiotic and biotic conditions such as climate, resource availability, competition or genotype identity can modify the allometric resource allocation pattern^14,31–34.

Plant community diversity is known to trigger changes in resource allocation patterns³⁵ through plastic responses of the constituent plants^36,37. Plastic responses of plants can contribute to niche differentiation processes, which in turn promote positive biodiversity–productivity relationships^38,39. In other words, plastic changes in resource allocation strategies in response to increasing plant diversity, such as a reduced reproductive effort due to relatively higher resource investment in vegetative plant parts with higher plant diversity, could diminish the biodiversity–seed yield relationship. However, this ecologically and economically very relevant question has, to our knowledge, not been scientifically addressed.

Understanding the abiotic and biotic factors concomitantly controlling the proportion of resources allocated to seeds is crucial for efforts to maintain or increase crop yields and to contribute to food security under a range of environmental and farming conditions. However, we lack an ecological understanding on how plant diversity, in interaction with the physical environment, influences the harvest index of the constituent species. For this study, we therefore selected eight annual grain crop species (oat, wheat, lentil, blue lupin, camelina, linseed, coriander and quinoa) commonly cultivated in Europe to determine their harvest index under varying species diversity levels, different climatic and soil fertility conditions and with locally adapted (i.e. home) versus foreign cultivars (i.e. away). To do this we conducted a common garden experiment replicated over two countries (Switzerland and Spain), two soil fertility levels (unfertilized and fertilised), two cultivars (Swiss and Spanish) and four plant diversity levels (i.e. isolated single plants, monocultures and 24 different 2- and 16 different 4-species mixtures) in a replicated fully factorial design.

Increasing crop diversity from monoculture to 2-species mixture increased seed yield by 3.4% in Spain and by 21.4% in Switzerland, while seed yield increases reached 12.7% and 44.3% from monoculture to 4-species mixture in Spain and Switzerland, respectively (Fig. 1).

Fig. 1. Seed yield response to crop diversity.

Average seed yield (g m^-2) of eight monocultures, 24 different 2- and 16 different 4-species mixtures planted with eight different annual crop species in 0.25 m² plots in Switzerland and Spain. Data are mean and 95% CI. n = 762 plots. See Supplementary Information Table 1 for the complete statistical analysis, and Extended Data Fig. 1 for vegetative biomass per plot and Extended Data Fig. 2 for seed yield and vegetative biomass of each species in monocultures and 2- and 4-species mixtures in Switzerland and Spain.

Even though crop diversity increased seed yield (Fig. 1), aboveground vegetative biomass increases with increasing crop diversity reached 25.8% and 46% in Spain and Switzerland, respectively, and were therefore 13.1% and 1.8% higher than the increases in seed yield (Fig. 2a). The reduced benefit of crop diversity on seed yield compared with vegetative biomass was due to a reduction in both types of mechanisms underlying diversity effects on yield, i.e. complementarity and sampling effects¹⁸ (Extended Data Fig. 3). When compared with the expected values for seed yield and vegetative biomass from the corresponding monocultures, it becomes apparent that seed yield is significantly lower than expected based on the corresponding data of vegetative biomass in Spain, but not in Switzerland (Fig. 2b).

Fig. 2. Crop plant diversity effects on seed yield and vegetative biomass.

Seed yield and vegetative biomass increases compared with monocultures averaged over 24 different 2- and 16 different 4-species mixtures, respectively. For absolute effect sizes we show the net effect (in g m^-2) (a) and for relative effect sizes the Relative Yield Total (b). n = 1274 aboveground biomass partitions (seeds vs vegetative) in plots of 0.25 m². Data are mean and 95% CI. See Supplementary Information Table 2 for the complete statistical analyses.

In line with these results at the plot level, we found at the individual plant level a clear trend towards reduced harvest index with increasing plant diversity (Fig. 3a). The harvest index in monocultures was higher than in mixtures — an effect only weakly dependent on species, country, ecotype and soil fertility (Fig. 3b; Supplementary Information Table 3). The strongest reductions in the harvest index from monocultures to 4-species mixtures where observed in Spain for oat (–22%), linseed (–9%), wheat (–4%), lupin (–4%) and coriander (–4%), and in Switzerland for lupin (–13%), lentil (–7%), linseed (–7%), wheat (–5%) and coriander (–3%). Finally, the harvest index was lower in 4-species mixtures than in 2-species mixtures (Fig. 3), except for locally adapted cultivars on fertilized soils (Extended Data Fig. 4).

Fig. 3. Harvest index of crop species in response to plant diversity and country.

The harvest index in response to plant diversity and country averaged over all species (a) and for each crop species separately (b). Data are mean and 95% CI. n = 4751 individuals. The harvest index of each species for each species combination is shown in Extended Data Fig. 5. See Supplementary Information Table 3 for the complete statistical analysis. Oat = Avena sativa, camelina = Camelina sativa, coriander = Coriandrum sativum, lentil = Lens culinaris, linseed = Linum usitatissimum, lupin = Lupinus angustifolius, quinoa = Chenopodium quinoa, wheat = Triticum aestivum.

The harvest index varied among species, being highest for legumes (i.e. L. culinaris (mean and 95% confidence interval): 0.60 [0.56, 0.63] and L. angustifolius: 0.57 [0.53, 0.62]), followed by herbs (i.e. C. sativum: 0.64 [0.61, 0.68], C. sativa: 0.55 [0.51, 0.59], L. usitatissimum: 0.51 [0.47, 0.55], C. quinoa: 0.49 [0.45, 0.53]), and lowest for cereals (i.e. A. sativa: 0.49 [0.48, 0.49], T. aestivum: 0.40 [0.37, 0.44]). The species-specific harvest index was also context-dependent and varied with ecotype and country and therefore with the home vs away environment. The harvest index was generally higher in Spain (0.56 [0.56, 0.57]) than in Switzerland (0.52 [0.51, 0.53]), which is consistent with previous studies which found that annual plants allocate relatively more resources to reproductive biomass than aboveground vegetative biomass under drought stress⁴⁰. Furthermore, the higher harvest index for legumes (lupin: +8%, lentil: +2%) and cereals (oat: +18%, wheat: +3.5%) in the home compared with the away environment provides evidence for the importance of local adaptation⁴¹ of crops for yield benefits (Extended Data Fig. 6).

A reduced harvest index in mixtures compared with monocultures was strongly linked to an increase in competition intensity, in particular for coriander, wheat, linseed, oat, lupin and lentil (Fig. 4a). This is in line with previous research demonstrating a drop of the harvest index with increasing planting density of crops⁴² and that in 2-species mixtures the subordinate species suffers a stronger decrease in the harvest index than does the dominant species⁴³. This mechanism might explain part of the diversity effects on the harvest index we see in our study, as we indeed see weaker responses of the community-level harvest index to an increase in species richness compared with the responses of individual species (Extended Data Fig. 7). Beyond that, reduced plant height (in particular wheat, linseed, lupin, oat, lentil and quinoa) together with increased leaf area (in particular lupin, linseed, lentil and oat) and specific leaf area [SLA] (in particular quinoa, oat and coriander) in mixtures compared with monocultures went along with a reduced harvest index (Fig. 4b–d). Finally, the harvest index was reduced when leaf dry matter content [LDMC] was higher in mixtures than in monocultures for linseed and quinoa, and when LDMC was lower in mixtures than in monocultures for lentil, coriander and lupin (Fig. 4e).

Fig. 4. Relationship of the harvest index of eight crop species with plant functional traits.

The difference in the harvest index of eight crop species in mixtures compared with monocultures as a function of differences in competition intensity (NIntC; a), vegetative plant height (b), leaf area (c), specific leaf area (SLA; d) and leaf dry matter content (LDMC; e) between mixtures and monocultures. Data are mean and 95% CI. n = 1637 species in plots of 0.25 m². See Supplementary Information Table 4 for the complete statistical analysis. Oat = Avena sativa, camelina = Camelina sativa, coriander = Coriandrum sativum, lentil = Lens culinaris, linseed = Linum usitatissimum, lupin = Lupinus angustifolius, quinoa = Chenopodium quinoa, wheat = Triticum aestivum.

The harvest index was highly responsive to the experimental treatments, including the different plant diversity levels, suggesting a plastic response of currently available crop plants to heterospecific neighbours in this trait. Specifically, the results demonstrate a deviation of resources away from reproduction towards the shoot with increasing neighbourhood plant diversity. This plastic response in resource allocation of crop plants in more diverse cropping systems compromises the yield benefits of crop mixtures. In the extreme case of oat in Spain, yield increases in mixtures compared with monocultures were reduced by 14 and 20% in 2- and 4-species mixtures, respectively, when compared with the expected yield benefits of oat calculated with the reproductive effort of oat in monocultures.

Our study demonstrates that beyond evidence for the benefits of intercropping for seed yield, growing currently available crop cultivars (commonly bred for high monoculture yield³) in mixtures does not result in the same amount of resources allocated to seed yield as in monocultures. As a caveat we point out that using small plots in our experiment might have led to an under- or, less likely, overestimation of effect sizes expected at field scale. Indeed, theoretical studies suggest that biodiversity–ecosystem functioning relationship should increase with spatial scale⁴⁴.

Despite the very large body of recent literature, we could not find any reference reporting diversity effects on reproductive effort in natural plant communities, indicating that perhaps these are rare⁴⁵. Therefore, it may well be that the current suite of crop cultivars does not represent a natural evolutionary constraint but rather represents unsuitable starting material to fully exploit the benefits of crop diversification for global food security. It is conceivable that less highly bred varieties would provide better starting material to breed crops for high mixture yield by avoiding a reduced harvest index in intercropping. In the same way as breeding for high monoculture yields was based on short-statured genotypes that engage less in intraspecific light competition in monocultures³, it may be possible to breed for high mixture yields if traits can be identified that increase complementarity among species above and below ground in mixtures. According to our results, these breeding programs may benefit from going back to locally adapted cultivars with traits that favour a higher harvest index in mixtures, such as tall plants with small leaves and low specific leaf area.

Methods

Study sites

The crop diversity experiment was carried out in outdoor experimental gardens in Zurich (Switzerland) and Torrejón el Rubio (Cáceres, Spain). The two sites had striking differences in climate and soil, which allowed us to use the different sites as a space-for-time design of traditional agricultural experiments conducted over multiple years to achieve robust results. Spain is Mediterranean semiarid while Switzerland is temperate humid. In Zurich, the garden was located at the Irchel campus of the University of Zurich (47.3961 N, 8.5510 E, 508 m a.s.l.). In Torrejón el Rubio, the garden was situated at the Aprisco de Las Corchuelas research station (39.8133 N, 6.0003 W, 350 m a.s.l.). During the growing season, the main climatic differences between sites are precipitation (572 mm in Zurich between April and August vs 218 mm in Cáceres between February and June) and daily average hours of sunshine (5.8 h in Zurich vs 8.4 h in Cáceres), but there is little difference in terms of temperature (average daily mean, min and max temperatures are 14.0 °C, 9.3 °C and 18.6 °C in Zurich vs 14.6 °C, 9.6 °C and 19.6 °C in Cáceres). All climatic data are from the Deutsche Wetterdienst (www.dwd.de) and are average values over the years 1961 to 1990.

Each experimental garden consisted of beds with square plots of 0.25 m² that were raised by 30 cm above the soil surface. In Switzerland, we had 554 plots spread over 20 beds of 1×7 m, with 26 to 28 plots per bed (Extended Data Fig. 8). In Spain, we had 624 plots spread over 16 beds of 1×10 m, with 38 to 40 plots per bed (Extended Data Fig. 9). The soil surface beneath the raised beds consisted of local bare soil, covered by a penetrable fleece. Each bed on top of the fleece was filled with 30 cm homogenised standard, but not enriched, local agricultural soil. The local soil in Switzerland was a neutral loamy soil consisting of 45% sand, 45% silt and 10% clay. Soil pH was 7.25, total C and N were 2.73% and 0.15%, respectively, and total and available P were 339.7 mg/kg and 56.44 mg/kg, respectively. The local soil in Spain was a slightly acidic sandy soil consisting of 78% sand, 20% silt and 2% clay. Soil pH was 6.39, total C and N were 1.02% and 0.06%, respectively, and total and available P were 305.16 mg/kg and 66.34 mg/kg, respectively. Therefore, compared with the soil in Switzerland, the soil in Spain was sandier and poorer in soil organic matter.

While the relatively small plot sizes allowed us to grow a large experiment under environmentally highly controlled but realistic outdoor conditions, some variables can suffer edge effects and interferences with neighbouring plots. However, such effects would likely increase residual variation more than between-treatment variation, because randomization was used to prevent confounding of between-plot interactions with treatments. In the only study we know of, biodiversity–productivity relationship in herbaceous communities were not affected by plots size⁴⁶ and a recent theoretical study showed that if anything biodiversity effects should increase with plot size⁴⁴. We therefore assume that effect sizes in our experiment, if anything, are probably rather conservatively estimated compared with studies using larger plot sizes.

Study species

The eight selected crop species were: Avena sativa (oat), Triticum aestivum (wheat), Lens culinaris (lentil), Lupinus angustifolius (blue lupin), Camelina sativa (camelina), Linum usitatissimum (linseed), Coriandrum sativum (coriander) and Chenopodium quinoa (quinoa). These species are important annual crops that can be cultivated in Europe. We only considered seed crops with similar growth requirements in terms of climate and growing-season length and with similar plant sizes to fit at least 40 individuals into the relatively small plots. The eight species belong to four phylogenetic groups, with two species per group. We had monocots [A. sativa (Poaceae) and T. aestivum (Poaceae)] vs dicots. Then, among the dicots, we differentiated between superasterids [C. sativum (Apiaceae) and C. quinoa (Amaranthaceae)] and superrosids. Among the superrosids, we finally differentiated between legumes [L. culinaris (Fabaceae) and L. angustifolius (Fabaceae)] and non-legumes [C. sativa (Brassicaceae) and L. usitatissimum (Linaceae)]. Each species was represented by two cultivars (hereafter called ecotypes), one local cultivar from Switzerland and another local cultivar from Spain (Supplementary Information Table 5). For cultivar selection we considered, whenever possible, traditional varieties with some inherent genetic variability within species.

Experimental design

The experimental design included a nested plant diversity treatment with four levels described by three nested contrasts. The first contrast differentiated between single control plants (between 4 and 10 replicates depending on species and country) and plant communities (i.e. contrast ‘Community’). The single-plant treatment was required to assess plant interaction intensity, i.e. the change in aboveground net primary productivity when growing as an isolated single plant compared with a plant with con- or heterospecific neighbours, in monocultures and mixtures respectively. Nested within the level plant communities, we differentiated between monocultures (two replicates for each species) and mixtures (i.e. contrast ‘Diversity’). Nested within the level of mixtures, we differentiated between 2-species mixtures (2 replicates of all 24 different species combinations consisting of two phylogenetic groups each) and 4-species mixtures (2 replicates of all 16 species combinations consisting of four phylogenetic groups each) (i.e. contrast ‘Species number’). To test for the context dependency of reproductive allocation patterns at different plant diversity levels, this setup was replicated at two levels of soil fertility (unfertilized control plots vs fertilized plots; factor ‘Fertilisation’). As fertiliser we used a formulated NPK fertiliser (ORGAMAX 7-12-7; Productos Agricolas MACASA S.L., Igualada, Spain), which was applied to whole beds. In the fertilised plots we applied 120 kg/ha N in the form of organic N (14.3%) and N-NH₄ (85.7%), 205 kg/ha P in the form of P₂O₅ and 120 kg/ha K in the form of K₂O divided over three fertilisation events, two events with 50 kg N and K and 85 kg P/ha applied one day before sowing and after tillering of wheat, and one event with 20 kg N and K and 34 kg P/ha during the flowering stage of wheat. The described experimental setup was further replicated for the Swiss and the Spanish ecotypes (i.e. factor ‘Ecotype’) both in Switzerland and in Spain (i.e. factor ‘Country’). The interaction between ‘Ecotype’ and ‘Country’ was assessed as additional factor ‘Home’, with two factor levels: ‘home’ representing Spanish cultivars in Spain and Swiss cultivars in Switzerland and ‘away’ representing the opposite combinations. With the present design, we preferred to assess the context-dependence of the diversity effect on crop plant growth through experimental manipulations of eight different environments (two countries × two soil fertility levels × two ecotypes) and replication of the experiment within each of these environments rather than replication over multiple years, as it is common for agronomic studies⁴⁷.

Experimental setup and data collection

In Spain, the seeds were sown between 2 and 4 February 2018 and in Switzerland between 4 and 6 April 2018. All the seeds were sown by hand at a standard sowing density for the corresponding crop species: 400 seeds/m² for cereals, 240 seeds/m² for superasterids, 592 seeds/m² for non-legume superrosids, and 160 seeds/m² for legumes. Total sowing density was the same for all species and for monocultures and mixtures, corresponding to a so-called substitutive mixture design⁴⁸. Sowing was conducted in four rows of 45 cm length per plot and an inter-row distance of 12 cm. Each row was composed of a single species (Extended Data Figs. 8 and 9). Sowing depth was 0.5 cm for C. sativa, 5 cm for L. culinaris and 2 cm for all other species. For the isolated single-plant treatment we placed five seeds in the center of the plot, randomly selected one plant approx. three weeks after germination and manually removed the spare individuals. Weeds were manually removed from all monoculture and mixture plots approx. 80 days after sowing, while weeds in the plots with isolated single plants were removed several times during the growing season to avoid competition of the single plants with the otherwise abundant weeds in these plots. No other interventions were made over the course of the experiment, e.g. no harrowing or pesticide application. Harvest was conducted for each species once it reached maturity and lasted in Spain between 15 June and 11 July except for C. quinoa, which was harvested between 26 July and 21 August. Harvest in Switzerland was between 11 and 13 July for C. sativa and for the rest of the species between 26 July and 5 September. The phenological development of individual species did not differ significantly between monocultures and mixtures, while some single plants tended to mature later (Extended Data Fig. 10). In each plot (except for isolated single plants) and for each species we randomly marked three individuals during the flowering stage (i.e. 6154 individuals). All the marked individuals were harvested separately and seeds (i.e. reproductive biomass) were separated from all other aboveground biomass, incl. stems, leaves and chaff (i.e. vegetative biomass). While seeds were air-dried, vegetative biomass was oven-dried at 80 °C for 48 h prior to weighing.

Data analyses

Plot-level yield responses to the experimental treatments were assessed using a linear mixed effects model with (1) country, ecotype and home vs away; (2) fertilisation, and (3) diversity and species number, and their interactions as fixed effects and species composition and bed ID as random effects. Plot-level yield as the total mass of all seeds produced in a plot was square-root transformed to meet assumptions of parametric statistics. Significance of each contrast⁴⁹ was assessed using type-I analysis of variance with Satterthwaite’s method.

In order to assess differences in biodiversity effects on vegetative biomass versus grain yield, we applied the additive partitioning method¹⁸ of biodiversity effects and calculated net effects, complementarity effects and sampling effects separately for vegetative biomass and grain yield. Differences in their responses to experimental treatments were tested with a linear mixed-effects model with net effect, complementarity effect or sampling effect as response variables and organ (shoot vs seeds), country, ecotype, home vs away, fertilisation, species number (2 vs 4) and their possible interactions as fixed effects. Bed ID and plot ID were included as random terms. The three response variables were square-root transformed to meet assumptions of parametric statistics. Significance of each contrast⁴⁹ was assessed using type-I analysis of variance with Satterthwaite’s method.

The harvest index was calculated for each sampled individual that produced seeds (i.e. 5107 individuals included, while 1047 individuals were excluded due to mortality, lack of mature seeds or missing data) as HI = seed biomass/(vegetative aboveground biomass + seed biomass). To detect the effects of (1) species, (2) country, ecotype and home, (3) fertilization, (4) species number (two- vs four-species) nested within diversity (monoculture vs mixture) nested within community (single individual vs community) and the possible interactions between these factors on the harvest index of the crops, we used a linear mixed-effects model and type-I analysis of variance. The harvest index was square-root-transformed to meet normality and homoscedasticity of variance assumptions. We included bed ID and plot ID as well as the species composition as random factors into the model.

In order to test for functional plant traits related to the harvest index of crops when neighbour diversity increased, we quantified differences in plant interaction intensity and plant functional traits between mixtures and monocultures and related them to the changes in the harvest index of plants from monoculture to mixture. Plant interaction intensity in the plots was calculated for each individual by means of the neighbour-effect intensity index with commutative symmetry (NIntC)⁵⁰.

NIntC = 2 × [ΔP/(P_-N+|ΔP|)], where P_-N is the aboveground net primary productivity (= vegetative biomass + seed yield) as a single plant without neighbours and ΔP is the difference in aboveground net primary productivity with (i.e. in a monoculture or a mixture) and without neighbours.

NIntC therefore quantifies the difference in aboveground net primary productivity of a plant individual in any monoculture or mixture compared with the average aboveground primary productivity of an individual of the same species and ecotype in the same country and soil fertility but growing in a plot of 0.25 m² as an isolated single plant without neighbours.

As plant traits we used vegetative plant height, leaf area, specific leaf area (SLA) and leaf dry matter content (LDMC). SLA is the ratio of leaf area over leaf dry weight (m² kg^-1) and LDMC is the ratio of leaf dry weight over leaf water-saturated fresh weight (mg g^-1). Together they reflect a fundamental trade-off in plant functioning between a rapid production of biomass (i.e. high SLA and low LDMC) and an efficient conservation of nutrients (i.e. low SLA and high LDMC)⁵¹, and the plant’s capacity of endurance and resistance in harsh environment^52–54. Vegetative plant height reflects plant’s ability to capture light energy in competition through relatively high growth rates^55,56. In a linear mixed-effects model we assessed the response of ΔHI_{mixture-monoculture} to ΔNIntC_{mixture-monoculture}, Δheight_{mixture-monoculture}, Δleaf area_{mixture-monoculture}, ΔSLA_{mixture-monoculture} and ΔLDMC_{mixture-monoculture} and their interactions with species. Bed and plot ID were included as random terms. Statistical significance of each factor was tested with type-III analysis of variance.

All analyses were conducted with R version 3.6.2⁵⁷. Reported figures, including means and confidence intervals are for estimated marginal means calculated using ggemmeans() in ggeffects ⁵⁸ and plotted with plot_model() in sjPlot ⁵⁹.

Extended Data

Extended Data Fig. 1. Vegetative biomass responses to crop diversity.

Average vegetative biomass (in g m^-2) of eight monocultures, 24 different 2- and 16 different 4-species mixtures planted with eight different annual crop species in 0.25 m² plots in Switzerland and Spain. Data are mean and 95% CI. n = 762 plots. The statistical analyses (type-I analysis of variance of a linear mixed model and significance tested with the Satterthwaite approximation method) show significant effects of country [F_(1,20.7) = 195.8, P = 5.23×10^-12], country × diversity [F_(1,677.5) = 8.1, P = 0.004], country × species number [F_(1,679.2) = 5.9, P = 0.015], and marginally significant effects of diversity [F_(1,45.1) = 3.7, P = 0.062] on vegetative biomass. See Extended Data Fig. 2 for seed yield and vegetative biomass data of each species in monocultures and 2- and 4-species mixtures in Switzerland and Spain.

Extended Data Fig. 2. Crop plant diversity effects on seed yield and vegetative biomass of each species.

Average seed yield (a) and vegetative biomass (b) per species (in g m^-2) of eight monocultures, 24 different 2- and 16 different 4-species mixtures planted with eight different annual crop species in 0.25 m² plots in Switzerland and Spain. Data are mean and 1 SE based on raw data. To facilitate interpretation, data were extrapolated so that the values reflect the area of 1 m² covered by the species in the community of interest. n = 2284 species in plots for seed yield and n = 2275 species in plots for vegetative biomass.

Extended Data Fig. 3. Crop plant diversity effects on seed yield and vegetative biomass.

Seed yield and vegetative biomass increases (in g m^-2) compared with monocultures averaged over 24 different 2- and 16 different 4-species mixtures, respectively. For the complementarity effect (a) n = 1274 biomass partitions (seed vs vegetative) in plots, for the sampling effect (b) n = 1181 biomass partitions (seed vs vegetative) in plots. Data are mean and 95% CI. See Supporting Information Table 6 for the complete statistical analyses.

In Switzerland, complementarity effects contributed 25% more than sampling effects to the net biodiversity effect on seed yield, while in Spain only sampling effects could be detected. Complementarity effects in Switzerland were 59% lower for seed yield than for vegetative biomass. Sampling effects were 70% and 83% lower for seed yield than for vegetative biomass in Spain and Switzerland, respectively.

Extended Data Fig. 4. Harvest index of crops in response to the Home vs Away, Fertilization, Country and Species number (2-vs 4-species mixtures) treatments.

Data are mean and 95% CI. n = 4751 individuals.

Extended Data Fig. 5. Harvest index of the eight crop species planted in communities of different species composition.

Species were abbreviated as: Avena sativa = Av, Triticum aestivum = Tr, Camelina sativa = Ca, Coriandrum sativum = Co, Lens culinaris = Le, Lupinus angustifolius = Lu, Linum usitatissimum = Li and Chenopodium quinoa = Qu. Data are mean and 1 SE. n = 4751 individuals.

Extended Data Fig. 6. Harvest index for eight crop species in their Home vs Away environment.

The harvest index quantifies the proportion of reproductive biomass, i.e. seed yield, from total aboveground biomass produced by the Spanish cultivars in Spain and the Swiss cultivars in Switzerland (Home) and vice versa (Away). Data are mean and 95% CI. n = 4751 individuals.

Extended Data Fig. 7. The harvest index at the community level in response to plant diversity and country.

Harvest index calculated at the community-level as the ratio between plot-level seed yield and plot-level aboveground biomass (i.e. the sum of aboveground vegetative biomass and seed yield), irrespective of species. Data shown are mean and 95% CI. n = 762 plots. The contrast of diversity vs mixture is marginally significant as a main effect [F_(1,44.9) = 3.05, P = 0.088], and significant in interaction with country [F_(1,675.1) = 4.96, P = 0.026], while the contrast between 2- and 4-species mixtures is not significant. Statistical tests were done on a linear mixed effects model with type-I analysis of variance and Satterthwaite approximation.

Extended Data Fig. 8. Layout of the experimental garden in Switzerland.

Monoculture and mixture communities are composed of four planting rows (divided by dashed lines and coloured strips), while plots with single plants are indicated in plain colour corresponding to the crop species. Plots (0.5 × 0.5 m) are delineated with a solid black frame. Beds (20 beds of 7 × 1 m each) are delineated with either a red (= fertilised) or green (= non-fertilised control) frame. Row numbers and column letters between the beds are used to identify each plot. Letters within the plots indicate the ecotype (S = Switzerland, E = Spain) of each crop species.

Extended Data Fig. 9. Layout of the experimental garden in Spain.

Monoculture and mixture communities are composed of four planting rows (divided by dashed lines and coloured strips), while plots with single plants are indicated in plain colour corresponding to the crop species. Plots (0.5 × 0.5 m) are delineated with a solid black frame. Beds (16 beds of 10 × 1 m each) are delineated with either a red (= fertilised) or green (= non-fertilised control) frame. Row numbers and column letters between the beds are used to identify each plot. Letters within the plots indicate the ecotype (S = Switzerland, E = Spain) of each crop species.

Extended Data Fig. 10. Phenological plant development for eight crop species as single plants, in monocultures and in mixtures in Switzerland and Spain.

Boxplots show median (black dot), upper and lower quartile (box), maximum and minimum values (whiskers) and outliers (open circles).

Three plant development stages for each of the eight crop species averaged over all treatments per country, per diversity level. The specific development stages are for oat: beginning of tillering, flowering (tip of inflorescence emerged from sheath), seeds with milk texture; for wheat: beginning of tillering, flowering, seeds milk ripe; for lentil: first true leaf unfolded, flowering, full seeds (seeds fill the pod cavities); lupin: stem elongation and bases of several leaves clearly separated from each other, flowering, green pods (septa split and filled with seeds); camelina: first true leaf unfolded, flowering, start of grain formation; linseed: first basal branches expanded, flowering, capsules expanded and seeds formed; coriander: first true coriander leaf unfolded, flowering, formation of green capsules; quinoa: first true leaves unfolded, flower buds in pyramids, start of grain formation.

Data availability statement

The data that support the findings of this study and the corresponding R-code are available on the public repository Zenodo (doi: 10.5281/zenodo.4750856).