Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Win-win opportunities combining high yields with high multi-taxa biodiversity in tropical agroforestry

## Abstract

Resolving ecological-economic trade-offs between biodiversity and yields is a key challenge when addressing the biodiversity crisis in tropical agricultural landscapes. Here, we focused on the relation between seven different taxa (trees, herbaceous plants, birds, amphibians, reptiles, butterflies, and ants) and yields in vanilla agroforests in Madagascar. Agroforests established in forests supported overall 23% fewer species and 47% fewer endemic species than old-growth forests, and 14% fewer endemic species than forest fragments. In contrast, agroforests established on fallows had overall 12% more species and 38% more endemic species than fallows. While yields increased with vanilla vine density and length, non-yield related variables largely determined biodiversity. Nonetheless, trade-offs existed between yields and butterflies as well as  reptiles. Vanilla yields were generally unrelated to richness of trees, herbaceous plants, birds, amphibians, reptiles, and ants, opening up possibilities for conservation outside of protected areas and restoring degraded land to benefit farmers and biodiversity alike.

## Results

### Biodiversity and vanilla yield

Comparing species richness and vanilla yield (kg/ha) in forest- and fallow-derived vanilla agroforests, we found that higher vanilla yields were not associated with a decrease in the overall species richness, nor the richness of endemic species, for trees, herbaceous plants, birds, amphibians, and ants. Notably, the analyzed relationship of yield with biodiversity is likely mediated by underlying variables such as management and land-use history. Moreover, the overall and endemic diversity of all taxa combined (i.e., their mean normalized richness) was also not related to vanilla yields at the plot level (Fig. 1; Supplementary Tables 1 and 2). We found a negative relationship between vanilla yield and butterfly species richness (Fig. 1G, estimate = −0.179, p-value < 0.001) and endemic butterfly species richness (Fig. 2G, estimate = −0.104, p-value = 0.043), a positive relationship with amphibians (Fig. 1E, estimate = 0.110, p-value = 0.045), and, depending on land-use history, an either positive (Fig. 2F, forest-derived, estimate = 0.289, p-value = 0.028) or negative (Fig. 2F, fallow-derived, estimate = −0.145, p-value = 0.016) relationship with endemic reptiles. Species richness of trees (Fig. 1B), reptiles (Fig. 1F) and mean normalized richness of all combined taxa (Fig. 1A) were higher in forest-derived than in fallow-derived vanilla agroforests (Supplementary Table 12). Similarly, when looking at endemics, land-use history mattered for species richness of trees, herbaceous plants, birds, and ants and mean normalized endemic richness, with higher values in forest-derived compared to fallow-derived vanilla agroforests (Fig. 2).

### Effects of land-use history on biodiversity of agroforests

The direction and magnitude of biodiversity responses at the local scale, i.e., the plot level, differed by land-use history and taxa. When compared with old-growth forest, we observed significant losses in species richness in forest-derived vanilla for birds (−38%), trees (−51%), and amphibians (−51%) as well as mean normalized richness of all taxa combined (−23%); whereas butterflies significantly gained species (+82%) (Supplementary Table 3). Herbaceous plants, reptiles, butterflies, and ants showed no significant difference between land uses (Fig. 1; Supplementary Tables 36). We also found significant losses in endemic species richness of amphibians (−57%), trees (−58%), birds (−69%), and mean normalized endemic richness (−47%) in forest-derived vanilla compared to old-growth forest. Gamma diversity was highest for old-growth forest across all taxa, except for the overall richness of herbaceous plants, butterflies, and ants as well as the endemic richness of butterflies (Supplementary Table 7). Compared with forest fragments, only butterflies significantly gained species (+122%) in forest-derived vanilla agroforests; losses or gains in species richness of all other taxa and of mean normalized richness were not statistically significant, also when focusing on endemic species (Supplementary Table 8).

When compared with fallows, we observed significant gains in overall species richness in fallow-derived vanilla agroforest only for trees (+149%) (Fig. 1; Supplementary Tables 46, S9). When looking at endemic species richness, we found significant gains for reptiles (+38%), and ants (+164%) as well as mean normalized endemic richness (+38%) when comparing fallow-derived vanilla agroforest to fallow (Fig. 2; Supplementary Tables 46, S9). Gamma diversity was higher for fallow-derived vanilla agroforest compared to fallow for all taxa, except for the overall richness of herbaceous plants, birds and butterflies as well as the endemic richness of trees (Supplementary Table 7). Notably, the proportion of the recorded species that are endemic to Madagascar varied strongly among the studied taxa, with 51% for trees, 20% for herbaceous plants, 61% for birds, 98% for amphibians, 74% for reptiles, 58% for butterflies, and 45% for ants, respectively. Taxa with high levels of endemicity are particularly prone to irrevocable biodiversity loss if land use negatively affects them. Sampling coverage across land-use types was on average 84% for all taxa, indicating a satisfactory sampling effort (Supplementary Table 10). In addition, the rarefaction curve for fallows based on species richness per sampling unit tended to reach an asymptote in fallows for amphibians, birds, reptiles, and trees (Supplementary Fig. 3). The rarefaction curves of ants, butterflies, and herbaceous plants did not reach an asymptote across all land-use types; thus, differences between land-use types may further increase if sampling effort is increased.

### Effects of land-use history on species composition inside agroforests

Across all taxa, species composition changed significantly from old-growth forest to forest-derived vanilla agroforest (Supplementary Fig. 4, Supplementary Table 11). Forest fragments and forest-derived vanilla agroforests differed significantly in species composition for herbaceous plants, butterflies, reptiles, amphibians, and marginally for ants. Comparing fallow to fallow-derived vanilla agroforest, we found significant changes in the composition of trees, herbaceous plants, birds, reptiles, butterflies, and ants (Supplementary Fig. 4). Amphibians did not differ in species composition between fallow and fallow-derived vanilla agroforest.

### Effects of environmental and management variables on yield

Vanilla yields varied widely and averaged at 105 ± SD 100 kg/ha (Supplementary Table 12). Vanilla yields increased with planting density (estimate = 2.901, SE = 0.415, p-value < 0.001) and vanilla vine length (estimate = 2.650, SE = 0.393, p-value < 0.001; Table 1; Supplementary Fig. 5 and Supplementary Tables 13 and 14). Moreover, vanilla yield tended to increase with labor input for hand pollination of vanilla flowers (estimate = 0.897, SE = 0.469, p-value = 0.069; Table 1). Importantly, vanilla yield was not related to canopy closure, slope, landscape forest cover, understory vegetation cover, soil characteristics, or elevation, suggesting a high intensification potential without the need for shade or vegetation removal (Supplementary Fig. 5). Furthermore, the age of vanilla plants did not influence vanilla yield. Lastly, vanilla yields also did not differ between fallow- and forest-derived vanilla agroforests (Tukey post-hoc test: estimate = 0.77, SE = 1.94, p = 0.691; Supplementary Fig. 6).

### Effects of environmental and management variables on biodiversity

To understand the effect of underlying yield-related management variables on biodiversity, we analyzed the relationship of species richness with four management and six environmental variables. We found trade-offs between yield-increasing variables (vanilla planting density and vanilla vine length) and overall and endemic species richness of trees and reptiles (Table 1, Supplementary Tables 1518, Supplementary Fig. 7) but win-wins with endemic species richness of herbaceous plants.

Firstly, agroforests with higher vanilla planting density had lower tree richness (estimate = −0.155 SE = 0.066, p-value = 0.019) and lower endemic tree richness (estimate = −0.337 SE = 0.096, p-value < 0.001) but higher endemic herbaceous plant species richness (estimate = 0.176, SE = 0.063, p-value = 0.005. Secondly, vanilla vine length was related to overall fewer tree species (estimate = −0.221, SE = 0.056, p-value = 0.001), endemic tree species (estimate = −0.553, SE = 0.104, p-value = 0.001) and reptile species (estimate = −0.0166, SE = 0.071, p-value = 0.019) as well as marginally fewer endemic reptile richness (estimate = −0.096, SE = 0.052, p-value = 0.068) at plot level. Notably, species richness of birds, amphibians, butterflies, and ants was driven by environmental and management variables unrelated to vanilla yields (Table 1, Supplementary Tables 1518, Supplementary Fig. 7). Apart from differences in species richness due to elevation, slope and soil characteristics, we found that canopy closure, a structural parameter strongly affected by farmers’ management decisions, was positively associated with species richness of trees (estimate = 0.284, SE = 0.082, p-value < 0.001), reptiles (estimate = 0.178, SE = 0.074, p-value = 0.016), endemic reptiles (estimate = 0.136, SE = 0.057, p-value = 0.017), endemic ants (estimate = 0.387, SE = 0.111, p-value < 0.010), and marginally with the species richness of endemic herbaceous plants (estimate = 0.145, SE = 0.078, p-value = 0.063). Additionally, a denser understory vegetation of shrubs and herbaceous plants was associated with a higher richness of endemic birds (estimate = 0.460, SE = 0.179, p-value = 0.010), but tended to reduce the richness of butterflies (estimate = −0.127, SE = 0.072, p-value = 0.077) and endemic butterflies (estimate = −0.109, SE = 0.049, p-value = 0.062). Landscape forest cover, mainly mediated by remaining forest fragments in the agricultural matrix, was positively associated with more species of trees (estimate = 0.270, SE = 0.074, p-value < 0.001), endemic trees (estimate = 0.827, SE = 0.107, p-value < 0.001), endemic herbaceous plants (estimate = 0.236, SE = 0.064, p-value < 0.001), endemic ants (estimate = 0.214, SE = 0.084, p-value = 0.011) and marginally reduced amphibian richness (estimate = −0.097, SE = 0.052, p-value = 0.062) in vanilla agroforests. Amphibian richness was lower on farms situated on steep slopes (estimate = −0.125, SE = 0.054, p-value = 0.021), whereas herbaceous plant richness was higher at higher elevations (estimate = 0.283, SE = 0.055, p-value < 0.001).

## Discussion

Here, we studied biodiversity–yield relationships in smallholder vanilla agroforests in Madagascar, a global biodiversity hotspot, with high pressure on the remaining old-growth forest and major sustainability challenges19. Vanilla is an important cash crop and a major export commodity of Madagascar that has the potential to lift tens of thousands of smallholder farmers out of poverty34. In contrast to common expectations of ecological-economic trade-offs3,35, we show that increasing yields within the current range of vanilla agroforestry management practices (productivity benchmark [reference point for an average maximum of yield] in Madagascar: 350 kg/ha36) are not generally associated with biodiversity losses. Moreover, our study highlights the great potential of vanilla agroforestry to restore the biodiversity value of degraded and fallow lands, which are prevalent in the main vanilla production region of northeastern Madagascar11,29. This potential is underlined by the fact that 70% of vanilla agroforests in the study region are already fallow-derived22. With targeted incentives for establishing vanilla agroforests on fallow land, vanilla agroforestry could further contribute to land restoration. By providing an alternative income it may also prevent the degradation of the last remnants of old-growth rainforest through shifting cultivation37. This is particularly important as our study shows that old-growth forests harbor high levels of species richness and unique species composition, which underlines their conservation value.

While vanilla yield increased with greater vanilla planting density and longer vanilla vines, we found trade-offs of biodiversity with yield-related variables for trees, endemic herbaceous plants, and reptiles. However, the species richness of the majority of taxa in vanilla agroforests was determined by non-yield-related management variables such as canopy closure and landscape forest cover, providing opportunities for smallholders to increase agricultural productivity. Importantly, these productivity increases do not come at the expense of biodiversity. Likewise, land-use history mattered for biodiversity but not for vanilla yields, indicating equal opportunities for profitable agroforestry on fallow land without further forest loss38. Higher landscape forest cover and higher canopy closure promoted endemic herbaceous plants and endemic species of trees, as well as endemic reptiles and ants in agroforests. Also, other studies from Madagascar have highlighted the importance of trees for biodiversity and identified the loss of canopy closure associated with species loss and community composition change across taxa39,40. Here, conservation of remaining forest fragments as well as farmer incentives for maintaining dense canopy structures is needed. Vanilla yield also tended to increase with labor input for hand pollination. While hand pollination is critical for achieving high vanilla yields41, this finding needs to be interpreted carefully, since it may also reflect an overall increase of pollination input with a high number of vanilla flowers present. Surprisingly, vanilla age was unrelated to vanilla yields, as the two are commonly positively related38.

Contrasting with the stable species richness of the majority of taxonomic groups, we found that agroforests with a high planting density of vanilla and long vanilla vines had reduced species richness of trees (overall and endemic) and reptiles as well astended to support fewer endemic reptile species. This indicates trade-offs between conservation and production goals, particularly for trees, which represent a keystone structure on which other species depend42. Doubling planting density from 3000 to 6000 vanilla plants/ha or vine length from 300 to 600 cm, corresponded to a decrease in tree richness by 27% or 23%, respectively. By contrast, almost tripling planting densities to 8500 plants/ha or quadrupling to 1200 cm vine length lowered tree species richness by 55% and 52%, respectively. Thus, intermediate increases in planting density and vanilla vine length can represent a compromise for tree conservation and vanilla production. High variations in planting density, as well as additional effects affecting yields (e.g., labor input), may be the drivers of tree species decreasing with more and longer vanilla plants, but not with increasing yields. Furthermore, our study highlights that tree richness strongly depends on landscape forest cover as well as canopy closure which can be achieved by conservation of remaining forest fragments as well as farmer incentives for maintaining old trees and dense tree canopies. While the mechanisms underlying the negative relationship of vanilla planting density with tree richness seem obvious, more research is needed on the mechanisms behind the negative relationships between vine length and tree as well as reptile diversity.

Our study supports findings from other major agroforestry systems such as cacao, demonstrating no relationship between cacao yield and multiple plants and animal taxa in Indonesia43 and Peru44. In contrast, a study on Cameroonian cacao agroforests found a negative relationship between ant richness and cocoa yields, suggesting trade-offs with biodiversity at high yield levels if shade trees are removed45. Vanilla smallholder agroforestry is not subject to similar trade-offs because the amount of shade vegetation is not related to vanilla yields38. However, vanilla has potential for intensification by increasing planting density and increasing vanilla vine length, which can be achieved by smallholder farmers without the need to reduce canopy closure. For example, doubling planting density from 3000 to 6000 vanilla plants/ha increases yields by 193%, that is, from 71 kg (2200 Euro/ha gross revenue) to 208 kg or 6400 Euro/ha at high vanilla prices of 201622. Similarly, doubling vine length from 600 to 1200 cm increases yields by 191% (66 kg 2000 Euro/ha to 192 kg 5900 Euro/ha). Notably, planting density was generally low (mean = 3284 plants/ha; SD = 1444; Supplementary Table 12) in our study region, compared to intensively managed vanilla systems elsewhere. In Mexico, vanilla planting densities reach up to 5000 plants/ha in monocultures and up to 15000–20000 plants/ha in shade houses46. Generally, vanilla yields in Madagascar still have a high intensification potential38.

Systematic reviews suggest that in about 80% of all studies, species richness in small-scale tropical agroforests is lower than in forests47. In line with our findings for trees, studies have highlighted that plants are more negatively affected by forest conversion than more mobile taxa like insects48,49. Nevertheless, agroforests can also support biodiversity levels similar to forests, if their transformation occurred recently and management remained extensive50,51. Vanilla agroforests, in contrast to coffee and cacao agroforestry systems, are generally extensively managed with highly variable yields, averaging at less than 500 kg green vanilla per hectare globally52. Malagasy vanilla is managed extensively and manually without inputs of fertilizers or pesticides22. The extensive management of vanilla as well as its setting in a diverse mosaic landscape with forest remnants contribute to their high biodiversity and species richness that for some taxa (e.g. butterflies, ants) can equal that of old-growth forest at the plot level53,54.

Our findings confirm recent calls that land-use history needs more consideration in agroforestry research and management15. The increase of biodiversity in fallow-derived vanilla agroforestry compared to fallows, particularly of endemics, presents considerable conservation opportunities in line with the goals of the UN Decade on Ecosystem Restoration (2021–2030) and the recent IPBES report5,55,56. In addition, we found compositional differences between fallow and fallow-derived vanilla agroforests for trees, herbaceous plants, birds, reptiles, butterflies, and ants, which may be explained by the tree regrowth and species colonizing fallow-derived vanilla agroforests. Once they are established, vanilla agroforests are unlikely to be transformed into other land-use types due to their high profitability57 and thus present a leverage point for breaking out of the shifting cultivation cycle that degrades much of the agricultural land in Madagascar57. Vanilla agroforests offer longlasting opportunities for biodiversity and tree stand structures to recover12,38, allowing associated biodiversity and ecosystem services to increase57 and species composition to be partially restored compared to fallow land54. Tree regrowth in fallow-derived agroforests is particularly valuable to supplement and connect the few remaining forest fragments across the agricultural matrix58. In contrast, fallow land under shifting cultivation experiences repetitive burning59, which limits the establishment and growth of trees11. However, ecological restoration also has to consider the value of fallow land to reconcile livelihood and conservation needs60. Fallow land can take different forms providing a wide array of benefits to people61. For example, right after subsistence rice production, fallows are often used for livestock grazing11. Indeed, the transformation of fallow land into agroforests can result in losses of provisioning ecosystem services (e.g., firewood, wild foods, timber)61. However, there is not yet a shortage of fallow land in northeastern Madagascar, and the loss of provisioning services through conversion to agroforests is readily offset by the associated benefits (e.g., cash crops, carbon storage)61,62. A heterogeneous landscape with multiple land uses is important to satisfy the needs of rural communities.

Madagascar has high levels of poverty, with around 65% of people depending on agriculture63,64. Inefficient land management and weak law enforcement are major challenges to biodiversity conservation and solutions to Madagascar’s biodiversity crisis are urgently needed20,65. To guide restoration measures and sustainable intensification, efforts need to be supported through well-designed policies and economic incentives. Farmer training should emphasize that agroforests on fallow land are as productive as forest-derived ones and that high canopy closure does not conflict with high yields. Contract farming with sustainability standards (e.g., Fair Trade, Rainforest Alliance) or compensations schemes (e.g., Payments for Ecosystem Services) may promote agroforestry on fallow land and the use of endemic trees. Such contractual arrangements may not only favor conservation but also farmers through greater income stability as well as guaranteed minimum and premium prices36,66.

Our study on Malagasy vanilla agroforestry is a prime example of how win-win solutions that combine high yields with high biodiversity can be achieved in tropical agriculture. While the last remnants of Madagascar’s old-growth rainforest need strict protection to conserve their unique biodiversity and species incompatible with agriculture, fallow-derived vanilla agroforestry can support the restoration of biodiversity and important ecosystem services within agricultural lands and provide a profitable alternative to further expansion into old-growth forest. Management strategies in vanilla agroforests allow both yields and biodiversity to be increased. Thereby, vanilla agroforestry opens up great opportunities for economically and ecologically sustainable land management in Madagascar, aligning with the UN ecosystem restoration goals in this outstanding tropical biodiversity hotspot.

## Methods

### Ethical statement

Ethics approval was obtained for this study from the ethics committee of the University of Goettingen (Chair: Prof. Dr. Peter-Tobias Stoll) under the reference number 17./04.22Wurz.

### Study area

All plots were situated in northeastern Madagascar in the SAVA region (Supplementary Fig. 1). The natural vegetation is tropical lowland rainforest, but deforestation rates are high30,67.

The region is globally and nationally one of the most biodiverse places with high levels of endemism17,68. Forest loss is mainly driven by slash-and-burn shifting hill rice cultivation58. The region is characterized by a warm and humid climate with an annual rainfall of 2255 mm and a mean annual temperature of 23,9 °C (mean value of 60 plots extracted from CHELSA climatology69). Vanilla is the main cash crop in the SAVA region, making Madagascar the main vanilla producer globally21,22. Vanilla prices have shown strong fluctuations over the past years, with a price boom between 2014 and 2019 triggering an expansion of vanilla agroforestry in the region22,23.

### Study design

We selected 10 villages based on the 60 villages selected within the Diversity Turn in Land Use Science project22 (Supplementary Fig. 1). We selected the villages based on the list of villages for our study region from official election lists which listed all villages within a fokontany individually22. Village boundaries, demographics, infrastructure were defined based on a rapid survey with the village chief. Among the 60 villages, we considered all villages without coconut plantations, with less than 40% water (river, sea, and lakes) to avoid a strong influence of water elements and with forest fragments and shifting cultivation present within a 2 km radius around the village. Two of these 17 villages overlapped within a 2 km radius of the villages, thus we randomly selected one of them, resulting in 14 villages. We visited these 14 villages in a randomized order and stopped after we found 10 villages which fulfilled the necessary criteria (all land-use types present, willing to participate). In each of the 10 villages, we selected three vanilla agroforests, one forest fragment, and two fallows. Overall, we studied 60 plots across 10 villages and 10 plots in one protected old-growth forest (Marojejy National Park). All plots had a minimum distance of 260 m and a mean minimum distance of 794 m (SD = 468 m) to each other. Plot elevation ranged between 10 and 819 m.a.s.l. (mean  = 205 m, SD = 213 m; Supplementary Table 20).

### Plot selection

In each of the 10 villages, we selected three vanilla agroforests with low, medium, and high canopy closure, respectively, covering a within village canopy cover gradient. To refine our vanilla agroforest classification, we used interviews with the plot owners to categorize all vanilla agroforests based on land-use history into fallow- and forest-derived agroforests15. Forest-derived vanilla agroforests are established within forest fragments, which have been manually thinned of dense understory vegetation. Fallow-derived vanilla agroforests are established on formerly slashed and burned plots, where vegetation has been cleared for hill rice production (shifting cultivation system locally called tavy). Out of our 30 vanilla agroforests, 20 vanilla agroforests were fallow-derived and 10 vanilla agroforests were forest-derived, roughly matching the proportion of fallow- and forest-derived vanilla agroforests across the study region (70% are fallow-derived vanilla agroforests, 27% are forest-derived vanilla agroforests and 3% of unknown origin22.

In addition to vanilla agroforests, we selected one forest fragment in each village. Forest fragments were located inside the agricultural landscape and were remnants of the once continuous forest; these fragments are frequently used for natural product extraction. Forest fragments have not been burned or clear cut in living memory, yet the ongoing resource extraction results in a much simplified stand structure and fewer large trees compared to old-growth forest12. Furthermore, we chose one herbaceous and one woody fallow in each of the 10 study villages. Both fallow types form part of the shifting hill rice production cycle and represent the fallow period at different stages after the crop production. Herbaceous fallows have been slashed and burned multiple times with the last cultivation cycle at the end of 2016, one year prior to the first species data collection in 2017, and thereafter left fallow11. The continuous succession of herbaceous fallows turns them into woody fallows with the domination of woody plants including shrubs, trees, and sometimes bamboo. Our 10 woody fallows have last burned 4–16 years before data collection. In this study, we combine both herbaceous and woody fallows into the category “fallow”. Generally, fallows occur in different forms in the study region. The characteristics of fallows depend on the frequency of past fires and the length of fallow periods in between crop cultivation11. Frequent burning results in a loss of native and woody species and a dominance of exotic species and grasses11. In later fallow cycles, fern species increasingly appear11.

Due to the commonly repeated slashing and burning, secondary forests are very rare in the study region. Shifting cultivation prevails in Madagascar70, because it is an important option for people to grow food because means for agricultural intensification are scarce. According to our baseline survey (performed in 60 villages in our study region), 90% of the interviewed farmers grow rice for subsistence in addition to growing vanilla22. Out of this sample, 64% of farmers grow rice in irrigated paddies and 26% of farmers use shifting cultivation.

We also studied 10 plots at two sites in Marojejy National Park, the only remaining, continuous old-growth forest at a low altitude in our study area71. We chose accessible old-growth forest plots with a minimum distance of 250 m from the forest edge. Five of the 10 old-growth forest plots were located in Manantenina Valley, the other five old-growth forest plots were situated in the eastern part of Marojejy National Park, called Bangoabe area. Illegal selective logging has occurred in some parts of the park. During our plot selection, we avoided sites with traces of selective logging.

### Land-use history classification

To collect information on the land-use history or farm history, interviews with farmers are common72,73. We did interviews with the plot owner. Questions on land-use history were binary (forest-derived or fallow-derived) and did not include information on the detailed land-use history (e.g. frequency of burning, past crop systems). Thus, we consider this selfreported data very reliable. The land-use categorization derived by farmers was confirmed by our visual plot inspections (forest-derived vanilla agroforests do have a quite distinctive vegetation structure compared to fallow-derived vanilla agroforests). Additionally, data on tree species composition and soil characteristics show evident differences between the categories and back up the binary land-use history categorization. Analysis of tree species composition showed that fallow- and forest-derived vanilla agroforests differ significantly in tree species composition12. Soil analysis (see Fig. S9) showed that our fallow-derived vanilla agroforests are associated with fertility-related variables such as an increase in calcium, pH, nitrogen, and phosphorus, which is common after slas-and-burn agriculture74,75.

### Plot design

We collected species data on plots with a radius of 25 m (1964 m2, 0.1964 ha). We established our circular plots in a homogeneous area of the land-use type or forest. Adjacent land uses were usually different because farmers generally own small-scale land with a mean size of 0.66 ha (mean size of agroforests). We assessed vanilla plant data (yield, vine length, vine age, planting density) on 36 vanilla pieds on each of 30 circular vanilla plots (Supplementary Fig. 8). We defined one vanilla pied (foot in French) as the combination of a vanilla vine and a minimum of one support tree. The 36 vanilla pieds were evenly selected in each of the circular plots based on a sampling protocol to ensure comprehensive and unbiased sampling. We chose vanilla pieds independent of age, length or health condition. We marked the 36 selected vanilla pieds per plot with a unique barcode to assess vanilla yield (April 2018) and other plant health variables on the same plant (not used in this study). However, for 37 vanilla pieds (out of a total of 1080 marked vanilla pieds), the barcodes were lost or unreadable and we selected a new plant closest to the original position (independent of age, length, or condition) and marked it with a new unique barcode. We measured the size of the vanilla agroforest by walking with the agroforest owner and a hand-held GPS device at the perimeter of the plot.

### Vanilla planting density

We counted each vanilla pied on each 25 m circular plot by dividing the plot in four-quarter segments. We calculated the area of each 25 m radius plot including slope correction and calculated vanilla planting density (vanilla pieds per hectare) by dividing the number of vanilla pieds by the slope-corrected plot area.

### Vanilla yield

We measured yield on 30 vanilla plantations (10 forest-derived vanilla plantations and 20 fallow-derived vanilla plantations); three in each of our 10 study villages. We measured vanilla yield on a total of 36 vanilla pieds between March and April 2018. We assessed the vanilla yield before harvest to ensure an accurate yield assessment due to two reasons. Firstly, vanilla pods are commonly harvested successively due to their differing pollination date and maturity requiring multiple visits over several weeks. Secondly, theft of vanilla pods is commonplace around harvest time. We, therefore, estimated the weight of the on-plant-hanging vanilla pods by measuring pod volume and relating this to a prior established volume–weight correlation. This is possible because vanilla pods only grow in length and width in the first 8 weeks of their development76. Our yield assessment consisted of one interview part with the plot owner and one measurement part. The interview part included questions about the occurrence of theft and early harvest on the plantation. During the measurement part, we assessed the number, diameter, and length of all vanilla pods. We measured vanilla pod length with a ruler starting at the junction of stem and pod until the tip of the pod without considering the bending of the pod. We measured the diameter at the widest part of the pod using a caliper. We firstly calculated pod volume based on the standard volume cylinder formula using the measured diameter (cm) and length (cm): V = πr2h.

Secondly, we calculated the weight (g) of each pod by using the linear regression equation (y = bx + a) of a weight–volume correlation of 114 vanilla pods from 114 different agroforests (weight, length, and diameter of these 114 green vanilla was assessed post-harvest in 2017). We calculated the weight of all measured pods of the harvest in 2018 based on the formula:

$${{{{{\rm{volume}}}}}}={{{{{\rm{\pi }}}}}}({{{{{\rm{diameter}}}}}}({{{{{\rm{mm}}}}}})/20)^\wedge 2\ast {{{{{\rm{length}}}}}}({{{{{\rm{cm}}}}}})$$

Here, we divided the pod diameter (mm) by 20 to obtain the radius and to transform millimeters to centimeters. Weight was defined as volume*0.5662 + 0.9699. No vanilla pods were stolen or already harvested on our 36 vanilla pieds and hence we did not need to account for it in our vanilla yield calculation.

### Vanilla vine length

We assessed vanilla vine length for all 36 vanilla pieds (same vanilla pieds as used for the yield assessment) on each plot by measuring the total length of the vine from the lowest to the highest part with a measuring stick. If the vanilla vine was looped on the support tree (= vanilla vine is hanging in multiple loops on the support tree), we measured from the top height of the looping of the vanilla vine until the lowest height of the vine. At the medium height of the vanilla vine, we counted the number of times the vanilla vine passed through. We calculated the total length of the liana by multiplying the maximum height of the vanilla vine by the number of times the vine passed through the middle. In some cases, the vanilla vine looped at two different heights, we thus considered the middle between the two looping heights as the top height. If vanilla vines grew on two different support trees, we considered them as one vanilla pieds if support trees were <30 cm apart. If the distance between both support trees exceeded 30 cm, we considered only the support tree with the most vines for the measurement.

### Pollination labor input

We performed a longitudinal survey with the plot owners of our 30 vanilla agroforests from October 2017 to October 2018. The questionnaire was pre-tested with 30 farmers in September 2017. All participants were trained by the four research assistants on how to use pictogram-supported questionnaires. Subsequently, a feedback workshop was held to adapt the pictograms and optimize the entire questionnaire. The pictograms had to be filled every day as a diary. Besides pollination labor input, we assessed the time spent on plantation establishment, planting, weeding, pruning, plantation safeguarding, harvesting, preparing (fermenting, drying, sorting), and selling of vanilla (not considered in this analysis). Every fortnight, trained assistants visited farmers to collect the diary questionnaires. Data entries that appeared unusual were verified with the farmers by the assistants. The diary questionnaires included questions on family labour input for pollination as well as other agricultural activities, such as weeding, harvesting, curing of vanilla, and others.

We decided to use pollination labour input as the only variable of labour input in our analysis for the following two reasons. Firstly, pollination is the most important and labour intensive part of the production77. Secondly, pollination has a defined time frame because vanilla only flowers between October and December22. Thus, the hours of pollination labour input are easy to disentangle and define for the farmers. In contrast, other tasks such as weeding, pruning, and planting often happen continuously and in parallel and are thus harder for the farmer to depict in working hours. Pollination is a labor-intensive activity as every vanilla flower is pollinated manually and requires agricultural know-how. Thus, the working hours can be related to the number of flowers or/and the worker’s know-how. We calculated pollination labour input per hectare by summing all working hours (Oct 2017–Oct 2018) and dividing it by the slope-corrected agroforest size. For three out of 30 vanilla agroforests pollination labour input was missing; we thus used the mean value of all 30 agroforests for the three missing values.

The household head who filled the pictograms received 10,000 Ariary (roughly 2.50€) per month. Pictograms are drawings made by a local artist which visually describe each of the working steps of vanilla cultivation (e.g. planting vanilla vine, weeding plantation, pollination). The amount of compensation was recommended as a reasonable compensation by locally experienced Malagasy project members. The sum was handed out by the local research assistants at the end of each month. All participants of the surveys were informed that participation is voluntary, that they can leave the survey anytime, and that all data is anonymized, i.e., no personal data will be published or shared with third parties. Guidelines of “Good Scientific Practice” by the University of Göttingen were adopted (adapted based on the recommendations of the codex for good scientific practice from the DFG, German Research Foundation; https://www.unigoettingen.de/en/good+scientific+practice/567647.html).

The interview methodology (i.e. informed consent by the test persons as well as the questionnaires) was evaluated by the ethics committee of the University of Goettingen (https://www.unigoettingen.de/en/534983.html) and complied with their principles of the Higher Education Act of Lower Saxony (NHG) and the constitutionally protected right of academic freedom (Reference number: 17./04.22-Wurz). Co-author H.H. designed the survey and trained the local assistants with B.F. until the assistants were able to enter data in a consistent and standardized manner. F.A., F. S. B., and the trained assistants conducted the interviews. The research assistants collected the data bi-weekly but visited the households weekly. F.A. and F. S. B. checked entry data bi-weekly and clarified false entries. Our research assistants were recruited mainly from the student population of the regional CURSA university center in Antalaha, which is located in the research area. The students speak the local Malagasy dialects. Questionnaire in the original language, used pictograms, and the reporting sheet for farmers are available on Open Science Framework: https://osf.io/z5uxs/?view_only=1bd699c5cda64023963e058254a33eec.

### Vanilla plant age

We assessed vanilla plant age by asking the farmer for each of the 36 vanilla pieds per plot. The Malagasy field researchers Evrard Benasoavina, Thorien Rabemanantsoa, and Gatien Rasolofonirina walked with the farmer to each vanilla plant (see the question “Ask farmer: How many years ago was the liana at this pied planted?” in uploaded original interview (yield assessment) on Open Science Framework: https://osf.io/wa2xn/?view_only=1bd699c5cda64023963e058254a33eec). Here, the age referred to the vanilla vine but not the support tree. In preparation for the farmer interviews, we prepared handouts in Malagasy language to inform the farmers about the scientific goals and the content of our data collection (see handout on Open Science Framework: https://osf.io/ndrxg/?view_only=1bd699c5cda64023963e058254a33eec).

### Canopy closure

We measured mean canopy closure at five subplots of our circular plots by taking hemispherical images with a Nikon D5100 camera, equipped with a Sigma Circular Fisheye (180°) 4.5 mm 1:2.8 lens. The camera was fixed on a tripod at 2.4 m height above vanilla support trees and understory vegetation. We selected the images with the best contrast of sky and vegetation using the histogram-exposure protocol and calculated canopy closure using a minimum thresholding algorithm78,79.

### Slope and elevation

We used the 30 m-resolution digital surface model “ALOS World 3D” by Japan Aerospace Exploration Agency (JAXA) to assess the mean slope and the mean elevation of each plot.

For all values, we applied slope correction80.

### Landscape forest cover

We calculated forest cover in a 250 m radius around each plot center based on binary forest cover data from 2017 with a 30 m resolution30 and the R-package raster81. To reliably identify forest, tree cover maps and satellite imagery with a tree cover threshold of 75% were combined30. Here, agricultural lands such as tree plantations are excluded by combining historical forest maps with up-to-date forest cover change maps82. We chose a 250 m radius as a compromise between mobile and immobile taxa.

### Understory vegetation cover

We estimated the vegetation cover (percentage woody and herbaceous cover) visually for the 0–2 m layers in % of five subplots on each plot (located in the plot center and at 16.6 m from the center in each cardinal direction) and calculated the mean understory vegetation cover per plot. We did not consider vanilla pieds in the estimation of the understory vegetation cover.

### Soil characteristics (PC1)

We took soil samples with a MacFadyen soil corer (5 cm diameter, 295 ml, 0–15 cm depth). We divided the plots into eight subplots, four subplots at 8.3 m distance to the plot center (inner area) and four subplots at 16.6 m distance to the plot center (outer area). In total, we collected four cores in the inner and outer area each, resulting in two mixed soil samples per plot. We stored each soil sample in a zip-lock bag until laboratory analysis. In the laboratory, we measured pH (H2O) with the fresh soil samples using 1:10 humus/water suspension after 24 h of equilibration. We measured pH (KCl) by adding 1.86 g KCl. Mean pH values by plot were calculated in logarithmic and back-transformed by using exponential. The remaining soil was dried at 70 °C and ground. We measured organic carbon (Corg) (mmol/g dry soil), total carbon (C) (mmol/g dry soil) and nitrogen (N) (mg/g dry soil) concentrations, and organic C-to-N ratio (mol/mol) by using the C/N elemental analyzer (Vario EL III, elementar, Hanau, Germany). Additionally, we determined effective cation exchange capacity in µmol/g dry soil of potassium (K), magnesium (Mg), calcium (Ca), iron (Fe), manganese (Mn), hydrogen (H), and aluminum (Al) by digesting oven-drier soil material in 65% HNO3 at 195 °C for 8 h. Samples were analyzed with inductively coupled plasma optical emission spectrometry (ICP-OES) (Optima 3000 XL, Perkin Elmer, USA). We calculated the total effective cation exchange capacity (µmol/gTB) by the sum of H, P, Mg, Ca, Fe, Mn, and Al and total base saturation (%) as the sum of K, Mg, and Ca. Extractable phosphorus (P) Resin (µmol/gTB) was measured using resin bags, which were placed in a soil–water suspension. Then, P was re-exchanged with NaCl and NaOH solutions and quantified colorimetrically after blue-dyeing.

Due to the high number of soil characteristics measured and possible multicollinearity, we calculated Spearman correlations using the cor and corrplot function from the corrplot R-package83. Based on collinearity, we excluded total effective cation exchange capacity, total base saturation, total C, organic C, effective cation exchange capacity of Mg, effective cation exchange capacity of pH percolate, and the effective cation exchange capacity of H percolate (Correlation matrix, Supplementary Fig. 9). We used the remaining variables, i.e. effective cation exchange capacity of Ca, K, Al and Mn, pH(KCl), total N, resin P, and organic C-to-N ratio to perform a principal component analysis (PCA) using the R-packages ggbiplot84 and factoextra85. The soil PC axis 1 (explained 45%) was mostly related to effective cation exchange capacity (µmol/gTB) of Ca, and K as well as pH(KCl), nitrogen (mmol/gTB), P(resin) (µmol/gTB) and the organic carbon–nitrogen ratio (mol/mol) while the soil PC axis 2 (explained 21%) was related to exchange capacity of Mg and Al and the Corg-to-N ratio (Supplementary Fig. 10). The coordinates of PC axis 1 were used as a proxy of soil characteristics for further analysis.

### Trees

We sampled trees on all land-use types except herbaceous fallows between September 2018 and January 201912. Access was denied to two fallow-derived vanilla plantations, resulting in 58 plots assessed overall (including 28 vanilla agroforests). We did a full inventory of all trees with freestanding stems of ≥8 cm diameter at breast height in each plot. This included trees, arborescent palms, herbs, and tree ferns but excluded lianas. We identified tree species with the help of a local tree expert (Chrysostome Bevao) and a taxonomic expert (Patrice Antilahimena) from Missouri Botanical Garden (Antananarivo, Madagascar). We derived information on origin and endemism for each species from the Tropicos Madagascar Catalog86. Voucher specimens are kept at the National Herbarium Tsimbazaza, Antananarivo (TAN) and the herbarium of the University of Mahajanga. Out of the 454 assessed species in this inventory, 276 (51%) were endemic to Madagascar.

### Herbaceous plants

We sampled herbaceous plants in eight subplots of 4 m2 each (32 m2 overall) between September 2018 and December 2019. In each subplot, we assessed vascular plant species without apparent wood at maturity40. We accounted for the possible seasonality variation of the plant phenology by sampling one village after another. In each village, we collected data on each land-use type except for the old-growth forests. Hence, the observations for each land-use type cover the study period along with the possible phenology variations. We determined each species’ endemism status from the Tropicos Madagascar Catalogue86. We stored all herbarium specimens at the Plant Biology and Ecology Department at the University of Antananarivo in Madagascar. From the 299 species assessed in this study, 59 species (20%) were endemic to Madagascar.

### Birds

We sampled birds during two 40 min point counts per plot with two observers per point count87 following a commonly used standardized method88. In all villages, we conducted one point count between September and December 2017 with co-author D.M. as the main observer and co-author R.A. as a second observer. The second point count was done between August and December 2018 with Eric Rakotomalala as the main observer and the co-authors D.M. or S.D. as the second observer. We exchanged the order of plot visits in the second year to minimize seasonal bias. For old-growth forest, we did all point counts in 2018; one in August 2018 with E.R. as the main observer and D.M. as the second observer, and a second count in December 2018 with E.R. as the main observer and S.D. as a second observer. E.R. and D.M. are both experienced birders and familiar with the encountered bird species due to previous fieldwork in Madagascar. S.D. and R.A. were trained prior to fieldwork to recognize calls and morphology of Malagasy bird species. The second observer was responsible for entering of data and supporting the identification. Main observers were responsible for spotting, hearing, and identifying the birds. On 63 plots we started one-point count around sunrise and one at least one hour after sunrise; on 7 plots this alteration was not possible due to logistical constraints. After arriving at the plot center, we waited for a minimum of three minutes to allow the birds to settle. We noted the conditions including rain (no rain, drizzle, light rain, heavy rain) and wind (Beaufort 1–1289) before each point count. We only started point counts under good weather conditions, meaning no rain and wind equal to or less than Beaufort 4. If weather conditions deteriorated for more than 10 min from the start of the point count, we aborted the point count and started again later or the next day under better conditions. For calculating bird species richness per plot, we disregarded observations only in flight and outside the 25 m radius of plots. We defined species as endemic if only occurring in Madagascar according to BirdLife species fact sheets90. Overall, we assessed 51 bird species of which 31 species (61%) were endemic to Madagascar.

### Amphibians and reptiles

We sampled amphibians and reptiles using repeated time-standardized search walks for 45 min by two observers91. We visited each plot both during the day and at night both during the driest (one nocturnal and one diurnal search between October and December 2017; one nocturnal and one diurnal search between August and December 2018) and the wettest period (one nocturnal and one diurnal search between January and April 2018 or in February 2019). We did so during the driest period and the wettest period. We systematically walked the circular plot in a zigzag pattern always starting from the West part toward North, East, South, and end in West to avoid counting twice the same individual during observation in one of the plots. We actively checked microhabitats to detect individuals hiding therein (e.g., individuals hiding under rocks, in leaf axils, tree barks, tree holes, leaf litter, or deadwood). When encountering an individual, we stopped the standardized search time and identified the individual92. We identified individuals based on morphological characteristics with the help of field guides93,94,95. We took DNA samples to determine the species for those individuals that proved difficult to identify using morphological characteristics only. To retrieve a DNA sample, we collected muscle or toe clips as tissue samples, conserved in 90% of alcohol. We stored DNA samples at the Evolutionary Biology laboratory at TU Braunschweig. We also took photos of specimens that we did not identify to species level (ventral, back, and flank view). Until release, we kept them in a ventilated bag to retain moisture. We released all specimens after completing the full-time-standardized search. We categorized endemic reptiles and endemic amphibians as species/morphospecies only occurring in the country of Madagascar96,97. We found in total 58 amphibian species of which 57 species (98%) were endemic. In terms of reptiles, we found 61 species of which 74% (45 species) were endemic to Madagascar (and 7 species without defined endemism level).

### Butterflies

We sampled butterflies with fruit traps and time-standardized netting between August and December 201853. We baited fruit traps with fermented bananas and deployed the cylindrical nets for 24 h. Before deployment, we fermented bananas for 48 h in an air-tight container.

On each plot, we installed a total of 8 fruit traps. We deployed four fruit traps at 16.6 m distance from the plot center in the four main cardinal directions and the other four fruit traps at 20 m distance from the center in the four intercardinal directions. We used fish lines covered with vaseline creme to hang the fruit traps. The Vaseline prevented ants to intrude on the fruit traps. In addition, we avoided any contact with branches on the fruit traps. We caught butterflies with a fruit trap with a 20 cm Cone Opening (90 cm long hanging 1.5 m above the ground. On plots without trees, we installed fruit traps on a support stick (in rice paddy and herbaceous fallow). During the 30 min time-standardized netting, we caught butterflies within an imaginary 2 m wide box to each side of the net while walking at a slow and steady speed in a zig-zag to equally cover the plot area. The net had a circular frame with a nylon mesh on a 1.5 m telescopic handle. We performed the timestandardized netting in dry and low-wind conditions only, either in the morning (8 a.m. to 12 p.m.) or in the afternoon (1 p.m. to 5 p.m.). We then collected and dried all captured butterflies and identified them to species level in the laboratory (moths excluded). Identification was done by Annemarie Wurz, David Lees and Sáfián Szabolcs. We categorized endemic butterflies as species/morphospecies only occurring in the country of Madagascar and updated with expert consultation by David Lees98. All identified specimens remain at the insect collection in the Department of Crop Sciences, section Agroecology, University of Göttingen, Germany. Overall, we found 84 butterfly species of which 49 species (58%) were endemic to Madagascar.

### Ants

We sampled ground-foraging ants using bait and pitfall traps54. We conducted the sampling in all villages between October and December 2017, and in the old-growth forest in August and December 2018. We established five sampling stations per plot: one at the plot center, and four at 16 m distance from the plot center; one in each cardinal direction. We then set bait and pitfall traps 10 m apart at each sampling station. We baited the bait traps using sardine and sugar on two white flat plastic plates with a diameter of 13 cm and placed the two plates about 5 cm apart. We left the baited traps for 30 min before collecting ants for 30 s. We buried the pitfall traps (plastic cups of 9 cm top diameter, 11 cm depth, and 6 cm bottom diameter) in the soil and filled them one-third with 70% alcohol and a few drops of soapy water. We emptied the pitfall traps after 48 h and identified ants to species/morphospecies level in the laboratory. Identification was done by Anjaharinony Rakotomalala, using available identification keys99,100,101. Cross-checking of the identification of the species was done with expert consultation by Jean Claude Rakotonirina (species of Leptogenys), Nicole Rasoamanana (species of Camponotus), and Manoa Ramamonjisoa (species of Tetramorium). We defined endemic ant species as those species only present in the country of Madagascar102. We stored voucher specimens at Madagascar Biodiversity Center, Antananarivo, Madagascar (MBC). We recorded in total 123 ant species of which 55 species (45%) were endemic to Madagascar.

### Statistical analysis

We performed all statistical analyses in R version 3.6.3103. To assess the overall species richness across all taxa, we calculated their mean normalized (endemic) richness. To ensure the equal weight of all taxa despite differences in the range of species richness per taxa, we normalized species richness measures with Min−Max Scaling, which is a common method in multi-taxa studies to compare species richness across different taxa and to calculate their overall richness104,105,106. Using this method, the data are linearly transformed to y = (x−min(x))/(max(x)−min(x)), where x is the set of observed values of species richness. As a result, the normalized species richness (y) then ranges between 0 and 1, with 0 referring to the lowest observed value and 1 to the highest observed value, respectively. We then calculated the average of the normalized values to obtain the mean normalized species richness, i.e., the overall richness of all taxa at the plot level. Our results were robust (no relationship between biodiversity and yield) independently of the metric used, i.e., normalized richness or multidiversity.

We investigated the relationship of species richness or normalized (endemic) richness with vanilla yield in 30 vanilla agroforests using glmmTMB models107 or a linear mixed-effects model108, respectively. We used glmmTMB due to its bigger flexibility (especially in the case of zero inflation) and its higher speed when using multiple fixed effects as well as random effects107. We treated vanilla yield (sqrttransformed) in interaction with land-use history (fallow vs. forest-derived) as an explanatory variable and site (village or old-growth forest site) as a random effect. We scaled and sqrt-transformed vanilla yield due to a few high-yielding plots inflating the data distribution.

We assessed the environmental and management-related covariates of species richness with a glmmTMB model107 and vanilla yield using a linear mixed effect model108 with yield sqrt-transformed. We tested for correlation among the covariates using the Spearman coefficient (Supplementary Fig. 11). For both models (species richness and vanilla yield) we used canopy closure, soil characteristics, slope, landscape forest cover, understory vegetation cover, elevation, planting density, pollination labour input, vanilla vine length, vanilla plant age as explanatory variables. We added site as a random effect and scaled all explanatory variables with the scale function. The function divides the centered columns (value per land-use type−mean value across land-use types) by their standard deviation across all land-use types. For the variable pollination labour input, data from three plots was missing. To avoid omitting these plots from the analysis, we performed a linear regression of available household labour with pollination labour/ha and found a good correlation between the two (p-value: 0.005, estimate = 52.82, SE = 17.38, Supplementary Fig. 12). Based on the number of household members (for which we have available data for the three missing plots), we then predicted expected pollination labour input for the three missing plots, using the estimated relationship of the two from the mentioned regression.

For all models assessing the environmental and management-related covariates of species richness, we used likelihood ratio tests using maximum-likelihood estimation109 to assess the statistical significance of individual variables. Specifically, we compared the full model versus a reduced model without the individual variable (single term deletion). The final model included all explanatory variables with residual deviance χ2 < 0.05 (“chi-squared value”) in the model comparison with the full model.

We assessed differences in species richness or normalized (endemic) richness between forest, forest fragment, forest-derived vanilla agroforests, fallow-derived vanilla agroforest, and fallows by fitting a glmmTMB model107 or a linear mixed-effects model108, respectively. We used a glmmTMB model for all analyses with count/discrete data as a response. For mean normalized richness, we used a linear mixed-effects model due to the continuous data. Here, we treated land-use type as an explanatory variable and site as a random effect. We tested if assumptions for normality and homogeneity of variances were met. If the residuals were homoscedastic, we used the glht function from the multcomp package110 applying Tukey’s all-pair comparisons with Bonferroni correction to assess differences in species richness between land-use types. If the residuals were heteroscedastic (withingroup deviation from uniformity significant with DHARMa quantile test111), we used the nonparametric Kruskal–Wallis test, followed by pairwise Wilcoxon test. For all our models we used simulation plots implemented in the DHARMa package to validate our model fit111. If our glmmTMB models were under- or over-dispersed, we changed the model family to compois (Conway–Maxwell Poisson distribution) and negative binominal (nbinom2), respectively. We assumed normal distribution for the models with mean normalized (endemic) richness as a response. To assess marginal and conditional R2, we used the delta method using the rsquaredGLMM function of the package MuMIn112.

We plotted all model results using RBase (Figs. 1, 2)103. The line inside the boxplot represents the median of each land-use type. We plotted scatterplots by using estimated regression lines from our fitted models and removed the vanilla agroforest type (forest- or fallow-derived vanilla agroforests) from the model if solely vanilla yield had a significant effect on species richness (Fig. 1 panels 5, 7). We extracted the model fits with the allEffects function from the effect package113,114. In all other cases (if no predictors were significant (e.g., Fig. 1 panel 3) or solely land-use history was significant as an additive or interactive effect (e.g., Fig. 1 panel 2) we retained the original model for plotting. In our graphs (Figs. 1 and 2), dashed horizontal lines are intercept-only linear models (lines are based on the mean of the distribution) and solid lines show statistically significant generalized linear regressions (P < 0.05). We show two colored lines as dashed lines if land-use history was significant as an additive term but no relationship of species richness with vanilla yield existed, or show them as solid lines if the effect of vanilla yield was moderated significantly by land-use history (Fig. 3).

To investigate differences in species composition among land-use types, we computed a permutational multivariate analysis of variance (PERMANOVA) using the adonis function of the vegan package115. Also, we used the pairwise.adonis function of the pairwiseAdonis package with false discovery rate correction to test for differences between land-use types116. Prior to PERMANOVA, we tested if homogenous dispersion existed among land-use types by using the betadisp function and permutest function of the vegan package (PERMDISP test)115. Our results show that heterogeneous dispersion significantly affected the differences in species composition for trees, birds, amphibians, butterflies, and ants (Table S14), thus differences are not only explained by the location of centroids but also by differences in dispersion between land-use types. We computed species composition by using non-metric multidimensional scaling (NMDS) with Jaccard dissimilarity distance. To compute the NMDS for amphibians, we excluded all plots where no amphibians occurred.

We generated sample-size-based rarefaction and extrapolation curves of species diversity of each taxon across land-use types to assess whether the curves reached an asymptote, indicating sampling completeness. To do so, we used ggiNEXT function of the iNext package117 with the following settings: type = 1, datatype = incidence_raw, q = 0 (species richness), endpoint = 20. Furthermore, we computed the sampling coverage in the percentage of each taxon across land-use types by using the function iNext with the same settings. We calculated gamma species diversity for each land-use type by summing up all species found across 10 plots and therefore excluded 10 out of 20 fallow-derived vanilla agroforests randomly (Table S18).

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

## Data availability

The data generated in this study have been deposited in the Open Science Framework database under accession code https://osf.io/j54fx/?view_only=1bd699c5cda64023963e058254a33eec.

## Code availability

The code used for this study has been deposited in the Open Science Framework database under accession code: https://osf.io/j54fx/?view_only=1bd699c5cda64023963e058254a33eec.

## References

1. Maxwell, S. L., Fuller, R. A., Brooks, T. M. & Watson, J. E. M. Biodiversity: the ravages of guns, nets and bulldozers. Nature 536, 143–145 (2016).

2. Clough, Y. et al. Land-use choices follow profitability at the expense of ecological functions in Indonesian smallholder landscapes. Nat. Commun. 7, 1–12 (2016).

3. Grass, I. et al. Trade-offs between multifunctionality and profit in tropical smallholder landscapes. Nat. Commun. 11, 1–13 (2020).

4. Macchi, L. et al. Trade-offs between biodiversity and agriculture are moving targets in dynamic landscapes. J. Appl. Ecol 57, 2054–2063 (2020).

6. Chazdon, R. L. Beyond deforestation: restoring forests and ecosystem services on degraded lands. Science (80-.) 320, 1458–1460 (2008).

7. Gann, G. D. et al. International principles and standards for the practice of ecological restoration. Restor. Ecol. 27, S1–S46 (2019). Second edition.

8. Schroth, G. et al. Agroforestry and Biodiversity Conservation in Tropical Landscapes (Island Press, 2004).

9. Tschora, H. & Cherubini, F. Co-benefits and trade-offs of agroforestry for climate change mitigation and other sustainability goals in West Africa. Glob. Ecol. Conserv. 22, e00919 (2020).

10. Santos, P. Z. F., Crouzeilles, R. & Sansevero, J. B. B. Can agroforestry systems enhance biodiversity and ecosystem service provision in agricultural landscapes? A meta-analysis for the Brazilian Atlantic. Forest. Ecol. Manag. 433, 140–145 (2019).

11. Styger, E., Rakotondramasy, H. M., Pfeffer, M. J., Fernandes, E. C. M. & Bates, D. M. Influence of slash-and-burn farming practices on fallow succession and land degradation in the rainforest region of Madagascar. Agric. Ecosyst. Environ. 119, 257–269 (2007).

12. Osen, K. et al. Land-use history determines stand structure and tree diversity in vanilla agroforests of northeastern Madagascar. Appl. Veg. Sci. 24, e12563 (2021).

13. Ocampo-Ariza, C. et al. Extinction thresholds and negative responses of Afrotropical antfollowing birds to forest cover loss in oil palm and agroforestry landscapes. Basic Appl. Ecol. 39, 26–37 (2019).

14. Warren-Thomas, E. et al. Rubber agroforestry in Thailand provides some biodiversity benefits without reducing yields. J. Appl. Ecol 57, 17–30 (2020).

15. Martin, D. A. et al. Land-use history determines ecosystem services and conservation value in tropical agroforestry. Conserv. Lett. 13, e12740 (2020).

16. Kremen, C. Reframing the land-sparing/land-sharing debate for biodiversity conservation. Ann. N. Y. Acad. Sci. 1355, 52–76 (2015).

17. Myers, N., Mittermeier, R. A., Mittermeier, C. G., da Fonseca, G. A. B. & Kent, J. Biodiversity hotspots for conservation priorities. Nature 403, 853–858 (2000).

18. Goodman, S. M. & Benstead, J. P. Updated estimates of biotic diversity and endemism for Madagascar. ORYX 39, 73–77 (2005).

19. Scales, I. R. Conservation and Environmental Management in Madagascar (Routledge, 2014).

20. Jones, J. P. G. et al. Last chance for Madagascar’s biodiversity. Nat. Sustain 2018, 9–11 (2019).

21. FAO. FAOSTAT Vanilla Production quantity. http://www.fao.org/faostat/en/#data/QC (2020).

22. Hänke, H. et al. Socio-economic, land use and value chain perspectives on vanilla farming in the SAVA Region (north-eastern Madagascar): The Diversity Turn Baseline Study (DTBS).Diskuss. No. 1806 July 2019, (2018).

23. Llopis, J. C. et al. Effects of protected area establishment and cash crop price dynamics on land use transitions 1990–2017 in north-eastern Madagascar. J. Land Use Sci. 14, 52–80 (2019).

24. Havkin-Frenkel, D. & Belanger, F. C. Handbook of Vanilla Science and Technology (WileyBlackwell, 2010).

25. Osewold, J. et al. Support trees in vanilla agroforests of Madagascar: diversity, composition and origin. Agrofor. Syst. 2022, 1–14 (2022).

26. Schwab, D., Wurz, A., Osen, K., Grass, I. & Tscharntke, T. Decreasing predation rates and shifting predator compositions along a land-use gradient in Madagascar’s vanilla landscapes. J. Applied Ecology 58, 360–371 (2020).

27. Chowdhury, F. I., Barua, I., Chowdhury, A. I., Resco de Dios, V. & Alam, M. S. Agroforestry shows higher potential than reforestation for soil restoration after slash-and-burn: a case study from Bangladesh. Geol. Ecol. Landscapes 6, 48–54 (2020).

28. van Vliet, N. et al. Trends, drivers and impacts of changes in swidden cultivation in tropical forest-agriculture frontiers: a global assessment. Glob. Environ. Change 22, 418–429 (2012).

29. Zaehringer, J. G., Hett, C., Ramamonjisoa, B. & Messerli, P. Beyond deforestation monitoring in conservation hotspots: analysing landscapemosaic dynamics in north-eastern Madagascar. Appl. Geogr. 68, 9–19 (2016).

30. Vieilledent, G. et al. Combining global tree cover loss data with historical national forest cover maps to look at six decades of deforestation and forest fragmentation in Madagascar. Biol. Conserv. 222, 189–197 (2018).

31. Kremen, C. & Merenlender, A. M. Landscapes that work for biodiversity and people. Science 362, eaau6020 (2018).

32. Broennimann, O. et al. Do geographic distribution, niche property and life form explain plants’ vulnerability to global change? Glob. Chang. Biol. 12, 1079–1093 (2006).

33. Chaudhary, A., Verones, F., De Baan, L. & Hellweg, S. Quantifying land use impacts on biodiversity: combining species-area models and vulnerability indicators. Environ. Sci. Technol. 49, 9987–9995 (2015).

34. ILO. Etat des lieux du travail des enfants dans la fillière vanille dans la region de la SAVA (International Labour Organisation of the United Nations, Antananarivo, 2011).

35. Perfecto, I. & Vandermeer, J. Biodiversity conservation in tropical agroecosystems: a new conservation paradigm. Ann. N. Y. Acad. Sci. 1134, 173–200 (2008).

37. Miller, D. C. et al. Forests, trees and poverty alleviation: policy implications of current knowledge. Policy Econ 131, 102566 (2021).

38. Martin, D. A. et al. Shade-tree rehabilitation in vanilla agroforests is yield neutral and may translate into landscape-scale canopy cover gains. Ecosystems 24, 1253–1267 (2021).

39. Scott, D. M. et al. The impacts of forest clearance on lizard, small mammaland bird communities in the arid spiny forest, southern Madagascar. Biol.Conserv. 127, 72–87 (2006).

40. Raveloaritiana, E. et al. Land-use intensification increases richness of nativeand exotic herbaceous plants, but not endemics, in Malagasy vanilla landscapes. Divers. Distrib. 27, 784–798 (2021).

41. Wurz, A., Grass, I. & Tscharntke, T. Hand pollination of global crops—asystematic review. Basic Appl. Ecol. 56, 299–321 (2021).

42. Tews, J. et al. Animal species diversity driven by habitat heterogeneity/diversity: the importance of keystone structures. J. Biogeogr. 31, 79–92 (2004).

43. Clough, Y. et al. Combining high biodiversity with high yields in tropicalagroforests. Proc. Natl Acad. Sci. USA 108, 8311–8316 (2011).

44. Jezeer, R. E., Santos, M. J., Verweij, P. A., Boot, R. G. A. & Clough, Y. Benefitsfor multiple ecosystem services in Peruvian coffee agroforestry systemswithout reducing yield. Ecosyst. Serv 40, 101033 (2019).

45. Bisseleua, D. H. B., Missoup, A. D. & Vidal, S. Biodiversity conservation,ecosystem functioning, and economic incentives under cocoa agroforestryintensification. Conserv. Biol. 23, 1176–1184 (2009).

46. Hernández, H. & Juan Lubinsky, P. Vanilla production in Mexico in vanilla. In Vanilla (eds Odoux, E. & Grisoni, M.) (CRC Press, 2010).

47. Scales, B. R. & Marsden, S. J. Biodiversity in small-scale tropical agroforests: areview of species richness and abundance shifts and the factors influencingthem. Environ. Conserv. 35, 160–172 (2008).

48. Steffan-Dewenter, I. et al. Tradeoffs between income, biodiversity, andecosystem functioning during tropical rainforest conversion and agroforestryintensification. Proc. Natl Acad. Sci. USA 104, 4973–4978 (2007).

49. Bhagwat, S. A., Willis, K. J., Birks, H. J. B. & Whittaker, R. J. Agroforestry: arefuge for tropical biodiversity? Trends Ecol. Evol. 23, 261–267 (2008).

50. Beukema, H., Danielsen, F., Vincent, G. & Hardiwinoto, S. & Van Andel, J.Plant and bird diversity in rubber agroforests in the lowlands of Sumatra,Indonesia. Agrofor. Syst 70, 217–242 (2007).

51. Harvey, C. A. & Villalobos, González J. A. Agroforestry systems conservespecies-rich but modified assemblages of tropical birds and bats. Biodivers.Conserv. 16, 2257–2292 (2007).

52. Odoux, E. & Grisoni, M. Medicinal and Aromatic Plants—Industrial Profiles. (CRC Press, Vanilla, 2010).

53. Wurz, A. et al. Land-use change differentially affects endemic, forest andopen-land butterflies in Madagascar. Insect Conserv. Divers. https://doi.org/10.1111/ICAD.12580 (2022)

54. Rakotomalala, A. A. N. A. et al. Tropical land use drives endemic versus exotic ant communities in a global biodiversity hotspot. Biodivers. Conserv. 30, 4417–4434 (2021).

55. IPBES. IPBES: Global assessment report on biodiversity and ecosystem servicesof the Intergovernmental Science-Policy Platform on Biodiversity and EcosystemServices (IPBES Secretariat, 2019).

56. Strassburg, B. B. N. et al. Global priority areas for ecosystem restoration. Nature 586, 724–729 (2020).

57. Martin, D. A. et al. Land-use trajectories for sustainable land systemtransformations: Identifying leverage points in a global biodiversity hotspot. Proc. Natl Acad. Sci. USA 119, e2107747119 (2022).

58. Zaehringer, J. et al. Revealing regional deforestation dynamics in North-Eastern Madagascar—insights from multi-temporal land cover changeanalysis. Land 4, 454–474 (2015).

59. Laney, R. M. Disaggregating induced intensification for land-change analysis:a case study from Madagascar. Ann. Assoc. Am. Geogr. 92, 702–726 (2002).

60. Fischer, J., Riechers, M., Loos, J., Martin-Lopez, B. & Temperton, V. M.Making the UN Decade On Ecosystem Restoration A Social–ecologicalEndeavour. Trends Ecol. Evol. 36, 20–28 (2021).

61. Zaehringer, J. G., Schwilch, G., Andriamihaja, O. R., Ramamonjisoa, B. & Messerli, P. Remote sensing combined with social-ecological data: theimportance of diverse land uses for ecosystem service provision in northeasternMadagascar. Ecosyst. Serv. 25, 140–152 (2017).

62. Soazafy, M. R. et al. Aboveground carbon stocks in Madagascar’s vanillaproduction landscape—exploring rehabilitation through agroforestry in thelight of land-use history. Glob. Ecol. Conserv 31, e01853 (2021).

63. The World Bank. Employment in Agriculture (% of Total Employment)(Modeled ILO Estimate). (The World Bank, Madagascar, 2021).

64. The World Bank. Poverty Headcount Ratio at \$1.90 a Day (% of Population). (The World Bank, Madagascar, 2019).

65. Rakotomanana, H., Jenkins, R. K. B. & Ratsimbazafy, J. Conservation Challenges for Madagascar in the Next Decade. In Conservation Biology. Voices fromthe Tropics. (eds. Raven, N. P. H., Sodhi, S. & Gibson. L.) 33–39 (2013).

66. Minten, B., Randrianarison, L. & Swinnen, J. F. M. Global retail chains andpoor farmers: evidence from Madagascar. World Dev 37, 1728–1741 (2009).

67. Moat, J. & Smith, P. P. Atlas of the Vegetation of Madagascar. (Royal BotanicGardens, Kew, 2007).

68. Brown, J. L. et al. Spatial biodiversity patterns of Madagascar’s Amphibiansand Reptiles. PLoS ONE 11, e0144076 (2016).

69. Karger, D. N. et al. Climatologies at high resolution for the earth’s land surfaceareas. Sci. Data 4, 1–20 (2017).

70. Curtis, P. G., Slay, C. M., Harris, N. L., Tyukavina, A. & Hansen, M. C. Classifying drivers of global forest loss. Science (80-.) 361, 1108–1111 (2018).

71. Goodman, S. M. A Floral and Faunal Inventory of the Parc national deMarojejy, Madagascar: with Reference to Elevational Variation. (Field Museumof Natural History, Chicago, USA, 2000).

72. Inoue, Y. et al. Traceability of slash-and-burn land-use history using opticalsatellite sensor imagery: a basis for chronosequential assessment of ecosystemcarbon stock in Laos. Int. J. Remote Sens. 28, 5641–5647 (2007).

73. Egeskog, A. et al. Actions and opinions of Brazilian farmers who shift tosugarcane an interview-based assessment with discussion of implications forland-use change. Land Use Policy 57, 594–604 (2016).

74. Béliveau, A. et al. Early effects of slash-and-burn cultivation on soilphysicochemical properties of small-scale farms in the Tapajós region,Brazilian Amazon. J. Agric. Sci. 53, 205–221 (2015).

75. Hölscher, D., Ludwig, B., Möller, R. F. & Fölster, H. Dynamic of soil chemicalparameters in shifting agriculture in the Eastern Amazon. Agric. Ecosyst.Environ. 66, 153–163 (1997).

76. Van Dyk, S. et al. Determining the harvest maturity of vanilla beans. Sci.Hortic. 168, 249–257 (2014).

77. Davis, E. W. Experience with growing vanilla (Vanilla planifolia). Acta. Hortic. 23–30 (1983).

78. Beckschäfer, P., Seidel, D., Kleinn, C. & Xu, J. On the exposure ofhemispherical photographs in forests. IForest —Biogeosciences 6, 228–237 (2013).

79. Beckschäfer, P. Batch Processing Hemispherical and Canopy Photographs with ImageJitle—User Manual. Chair of Forest Inventory and Remote Sensing,Georg-August-Universität Göttingen, Germany. (2015).

80. Aerospace Exploration Agency. ALOS Global Digital Surface Model. ALOS World 3D—30m (AW3D30) (Aerospace Exploration Agency). 201581. https://rspatial.org/raster (2019).

81. Hijmans, R. J. et al. Package ‘raster’ https://rspatial.org/raster (2019).

82. Harper, G. J., Steininger, M. K., Tucker, C. J., Juhn, D. & Hawkins, F. Fiftyyears of deforestation and forest fragmentation in Madagascar. Environ.Conserv. 34, 325–333 (2007).

83. Wei, T. & Simko, V. R package ‘corrplot’: visualization of a correlation matrix. https://cran.r-project.org/web/packages/corrplot/corrplot.pdf (2017).

84. Vu, V. Q. ggbiplot: A ggplot2 based biplot. https://github.com/vqv/ggbiplot (2011).

85. Kassambara, A. & Mundt, F. factoextra: Extract and Visualize the Results of Multivariate Data Analyses. R Package Version 1.0. 5. R package version (2020).

87. Martin, D. A. et al. Bird diversity and endemism along a land-use gradient in Madagascar: the conservation value of vanilla agroforests. Biotropica 53, 179–190 (2021).

88. Bibby, C., Burgess, N., Hillis, D., Hill, D. & Mustoe, S. Bird Census Techniques. Elsevier (2000).

89. Beer T. (2013) Beaufort Wind Scale. In Encyclopedia of Natural Hazards. Encyclopedia of Earth Sciences Series. (ed. Bobrowsky, P.T.) (Springer, Dordrecht, 2013).

90. BirdLife International. BirdLife International: Data Zone. http://datazone.birdlife.org/home (2018).

91. Fulgence, T. R. et al. Differential responses of amphibians and reptiles to land‐use change in the biodiversity hotspot of north‐eastern Madagascar. Animal conservation https://doi.org/10.1101/2021.03.18.435920 (2021).

92. Glaw, F. & Vences, M. Plethodontohyla guentheri, a new montane microhylidfrog species from northeastern Madagascar. Mitteilungen aus dem Mus. f.üR.Naturkd. Berl. – Zool. R 83, 33–39 (2007).

93. Ratsoavina, F. M. et al. Finaritra! A splendid new leaf-tailed gecko (Uroplatus)species from Marojejy National Park in north-eastern Madagascar. Zootaxa 4545, 563–577 (2019).

94. Rakotoarison, A., Scherz, M., Glaw, F. & Describing, J. K. the smallermajority: Integrative taxonomy reveals twenty-six new species of tinymicrohylid frogs (genus Stumpffia) from Madagascar. Vertebr. Zool. 67, 271–398 (2017).

95. Glaw, F. & Vences, M. A field guide to the amphibians and reptiles of Madagascar. (Vences & Glaw, Köln, 2007).

96. The Reptile Database. The Reptile Database. Uetz, P. Freed, P. Hošek, J. http://www.reptiledatabase.org (2019).

97. AmphibiaWeb. AmphibiaWeb https://amphibiaweb.org (University ofCalifornia, Berkeley, CA, USA, 2019).

98. Lees, D. C., Kremen, C., Raharitsimba, T., Goodman, S. M. & Benstead, J. P. Classification, diversity, and endemism of the butterflies (Papilionoidea andHesperioidea): a revised species checklist. in The Natural History ofMadagascar (eds Goodman, S. M. & Benstead, J. P.) (University of ChicagoPress, 2003).

99. Fisher, B. L. & Smith, M. A. A revision of Malagasy species of Anochetus Mayrand Odontomachus Latreille (Hymenoptera: Formicidae). PLoS ONE 3, e1787 (2008).

100. Bolton, B. & Fisher, B. L. The Madagascan endemic myrmicine ants related toEutetramorium (Hymenoptera: Formicidae): taxonomy of the generaEutetramorium Emery, Malagidris nom. n., Myrmisaraka gen. n., Royidrisgen. n., and Vitsika gen. n. Zootaxa 3791, 1–99–1–99 (2014).

101. Rakotonirina, J. C. & Fisher, B. L. Revision of the Malagasy ponerine ants ofthe genus Leptogenys Roger (Hymenoptera: Formicidae). Zootaxa 3836, 1–163–1–163 (2014).

102. AntWeb. Version 8.77.4. California Academy of Science. https://www.antweb.org (2020).

103. R Core Team. A Language and Environment for Statistical Computing (R CoreTeam, 2020).

104. Tittensor, D. P. et al. Global patterns and predictors of marine biodiversityacross taxa. Nature 466, (2010).

105. Collen, B. et al. Global patterns of freshwater species diversity, threat andendemism. Glob. Ecol. Biogeogr 23, 40–51 (2014).

106. Van Elsas, J. D. et al. Microbial diversity determines the invasion of soil by abacterial pathogen. Proc. Natl Acad. Sci. USA 109, 1159–1164 (2012).

107. Brooks, M. E. et al. glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. The R journal 9, 378–400 (2017).

108. Bates, D., Martin, M., Bolker, B. & Walker, S. Fitting Linear Mixed-Effects Models Using. {lme4}. J. Stat. Softw. 67, 1–48 (2015).

109. Zuur, A. F., Ieno, E. N., Walker, N. J., Saveliev, A. A. & Smith, G. M. Mixed effects modelling for nested data. In Mixed effects models and extensions in ecology with R. Vol. 574, 101–142 (Springer, New York, NY, 2009).

110. Hothorn, T., Bretz, F. & Westfall, P. Simultaneous inference in generalparametric models. Biometr. J 50, 346–363 (2008).

111. Hartig, F. DHARMa: Residual Diagnostics for Hierarchical (Multi-Level/Mixed) Regression Models. R Package Version 0.2.7. http://florianhartig.github.io/DHARMa/ (2020).

112. Barton, K. MuMIn: Multi-Model Inference. http://r-forge.r-project.org/projects/mumin/ (2009).

113. Fox, J. & Weisberg, S. An R companion to applied regression. Sage publications (2018).

114. Fox, J. & Weisberg, S. Visualizing fit and lack of fit in complex regressionmodels with predictor effect plots and partial residuals. J. Stat. Softw. 87,(2018).

115. Oksanen, J. et al. vegan: Community Ecology Package. https://github.com/vegandevs/vegan (2020).

116. Arbizu, P. M. pairwiseAdonis: Pairwise multilevel comparison using adonis. Rpackage version 0.0, Volume 1. (2017)

117. Hsieh, T. C., Ma, K. H. & Chao, A. iNEXT: Interpolation and extrapolation for species diversity. R package version 2.0.12 (2016).

## Acknowledgements

We thank all village leaders, village facilitators, and regional authorities who supported, communicated, and permitted our research. Special thanks go to all plot owners and their family members who allowed us to work on their fields and committed their time to facilitate our research. We thank Madagascar National Parks and the Ministry of Environment and Sustainable Development for granting us access to sites and allowing us export of samples. We express our appreciation to Rainer Marggraf and Andrea D. Bührmann as project leaders. Thanks go to Theudy Alexis, Patrice Antilahimena, Evrard Benasoavina, Claudine Bemamy, Jean Chrysostome Bevao, Ronik Botra, Joel Arnaud Harisaina, Dietrich Hertel, David Lees, Adriane März, Johannes Osewold, Thorien Rabemanantsoa, Dorah Ramaharobandro, Manoa Ramamonjisoa, Julien Randriampenomanana, Cédric Randrianantenaina, Eric Rakotomalala, Nicole Rasoamanana, Nantenaina Herizo Rakotomalala, Gatien Rasolofonirina, Joel Razafinantenaina, Dominik Schwab, Jean Claude Rakotonirina, Szabolcs Sáfián, Guillaume Velotody, Miguel Vences, and Maria S. Vorontsova who were indispensable to data collection, sample processing and sample identification. We collected data under research permits Nos. 100/17/MEEF/SG/DGF/DSAP/SCB.Re, 163/17/MEEF/SG/DGF/DSAP/SCB.Re, 18/18/MEEF/SG/DGF/DSAP/SCB.Re, and 254/18/MEEF/ SG/DGF/DSAP/SCB.Re granted by the Ministry for Water, Ecology and Forest (MEEF), Antananarivo. This study was financially supported by the Niedersächsisches Vorab of Volkswagen Foundation as part of the research project ‘Diversity Turn in Land Use Science’ (Grant number 11-76251-99-35/13 (ZN3119)) and German Academic Exchange Service (DAAD) within the ‘Partnerships for Supporting Biodiversity in Developing Countries’ initiative (Project No. 57449386). NG-R thanks the Dorothea Schlözer Postdoctoral Program of the Georg-August-Universität Goettingen for their support. Portions of this paper were developed from a dissertation chapter of Annemarie Wurz and Dominic A. Martin.

## Funding

Open Access funding enabled and organized by Projekt DEAL.

## Author information

Authors

### Contributions

A.W., T.T., I.G., D.A.M., K.O., E.R., F.A., T.R.F., A.A.N.A.R., M.R.S., A.A., J.B., H.H., D.H., H.K., B.R., H.L.T.R., F.M.R., L.H.R. conceived the idea; A.W., D.A.M. and K.O. developed the yield assessment protocol and collected the yield data; D.A.M. prepared yield data; A.W., D.A.M., K.O., E.R. T.R.F., M.R.S., A.A.N.A.R., R.R., R.A., and S.D. collected and processed species data; F.S.B. coordinated logistics; F.A., F.S.B., H.H. collected labour input data; A.W. analyzed and visualized the data; N.G.-R., A.A.N.A.R. and E.R. assisted analysis; A.W. led the writing of the manuscript. All authors contributed to the writing and gave final approval for publication. Authors are ordered (1) in alphabetic order for contributing Ph.D.s (FA-MRS) and (2) in alphabetic order for contributing project members, PostDocs and Professors (R.A. and L.H.R.).

### Corresponding author

Correspondence to Annemarie Wurz.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

## Peer review

### Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Wurz, A., Tscharntke, T., Martin, D.A. et al. Win-win opportunities combining high yields with high multi-taxa biodiversity in tropical agroforestry. Nat Commun 13, 4127 (2022). https://doi.org/10.1038/s41467-022-30866-8

• Accepted:

• Published:

• DOI: https://doi.org/10.1038/s41467-022-30866-8