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

Agricultural expansion and intensification are major drivers of biodiversity loss, and increases in productivity often come at the expense of biodiversity. Finding solutions to these ecological-economic trade-offs is urgent, particularly in rapidly transforming tropical landscapes. Restoration, including through agroforestry, is promoted as part of the UN Decade on Ecosystem Restoration, yet outcomes depend strongly on land-use history—whether agroforests replace forests or are established on open/degraded land. Agroforestry’s biodiversity value must also be considered alongside productivity, because low-yield systems can drive more land conversion. Madagascar, a global biodiversity hotspot with high endemism and severe poverty, has rapidly expanded smallholder vanilla agroforestry, established either by thinning forests or on fallow land. The study investigates biodiversity–yield relationships in smallholder vanilla agroforests in northeastern Madagascar across seven taxa, explicitly considering land-use history (forest- vs fallow-derived), and differentiating total and endemic species richness. The goals are to assess whether higher vanilla yields trade off with biodiversity, quantify biodiversity differences relative to old-growth forest, forest fragments, and fallows, and identify environmental and management drivers of yields and species richness to inform biodiversity-friendly, profitable agroforestry.

The paper situates its research within evidence that agricultural intensification tends to reduce biodiversity and that outcomes of agroforestry depend on land-use history. Prior studies often overlooked land-use origin when assessing biodiversity in tropical agroforests. Restoration via agroforestry can recover biodiversity and ecosystem services on degraded lands, but agroforestry may cause losses if replacing forests. Productivity is central to conservation outcomes due to land-sparing/sharing dynamics. Evidence from other agroforestry systems (e.g., cacao) shows mixed biodiversity–yield relationships, with some systems showing no yield-biodiversity relationship and others showing trade-offs when shade is reduced. Madagascar’s landscapes have high endemism and ongoing deforestation driven by shifting cultivation, with expanding smallholder vanilla systems that can be established on fallows, potentially aiding restoration. The study builds on calls to integrate land-use history and multi-taxa assessments, and to distinguish endemic species due to their vulnerability.

Ethics approval was obtained from the University of Göttingen ethics committee (Ref. 17./04.22-Wurz). Study area: SAVA region, northeastern Madagascar, a warm, humid tropical lowland rainforest area with high endemism and high deforestation rates. Study design: 60 plots were established across 10 villages (each with 3 vanilla agroforests, 1 forest fragment, 2 fallows) plus 10 plots in old-growth forest at Marojejy National Park, totaling 70 plots. Vanilla agroforests (n=30) were classified by land-use history via owner interviews and field verification into forest-derived (n=10) and fallow-derived (n=20). Plots were at least 260 m apart (mean 794 m) and spanned 10–819 m elevation. Plot layout: Circular plots of radius 25 m (0.1964 ha). In vanilla plots, 36 vanilla pieds (vine + ≥1 support tree) were randomly selected and barcoded per plot for plant/yield measurements. Vanilla measurements: Planting density (pieds/ha) was counted within plots and area slope-corrected. Vanilla yield was measured in 2018 on 36 pieds per agroforest (n=30 agroforests) before harvest by measuring pod length and diameter, computing volume (V = πr^2h; volume = (diameter(mm)/20)^2 × length(cm)), and converting to weight via a pre-established weight–volume regression (weight = 0.5662 × volume + 0.9699). Vine length per pied was measured, including loop counts to derive total length. Pollination labor input (hrs/ha) was recorded with pictogram-supported daily diaries (Oct 2017–Oct 2018) collected biweekly; for three missing entries, values were predicted from household size via regression. Vanilla plant age was reported by farmers for each pied. Environmental and structural covariates: Canopy closure was quantified from hemispherical photographs (five subplots per plot) using a minimum thresholding algorithm; slope and elevation came from ALOS World3D (30 m), with slope correction; landscape forest cover (% within 250 m radius) from 2017 30 m binary forest maps combining historical and recent datasets; understory vegetation cover (%) in 0–2 m layer visually estimated at five subplots; soils were sampled (0–15 cm) at eight cores pooled into two composites per plot, with pH(H2O, KCl), Corg, total C and N, C:N, exchangeable cations (K, Mg, Ca, Fe, Mn, H, Al), base saturation, CEC, and resin-extractable P measured. To reduce multicollinearity, a PCA was run on selected soil variables; PC1 (45% variance; related to Ca, K, pH(KCl), N, resin-P, C:N) was used as the soil characteristics proxy. Taxon sampling: Trees (DBH ≥ 28 cm) were fully inventoried in all land uses except herbaceous fallows across 58 plots (28 agroforests); identifications used local and herbarium experts; endemism from Tropicos Madagascar Catalogue. Herbaceous plants were sampled in eight 4 m² subplots per plot (total 32 m²) across villages, with specimens vouchered at University of Antananarivo. Birds were surveyed with two 40-min point counts per plot (two observers, standardized weather/time) across 2017–2018; endemism per BirdLife. Amphibians and reptiles were surveyed by repeated standardized searches (45 min, two observers) day and night in dry and wet seasons; difficult identifications verified via DNA; endemism defined as Madagascar-only occurrence. Butterflies were sampled by eight fermented banana-baited fruit traps (24 h) and 30-min standardized netting per plot (Aug–Dec 2018); endemism via experts. Ants were sampled with bait and pitfall traps at five stations per plot; specimens identified to species/morphospecies; endemism defined as Madagascar-only occurrence. Statistics: Overall and endemic multi-taxa richness were computed as mean normalized richness across taxa via min–max scaling (0–1). Relationships between species richness (individual taxa and mean normalized) and sqrt-transformed vanilla yield were modeled using glmmTMB (and linear mixed-effects for normalized metrics), with yield × land-use-history interaction and site as a random effect. Environmental/management drivers of yield were analyzed with linear mixed-effects models (yield sqrt-transformed); species richness drivers with glmmTMB including canopy closure, soil PC1, slope, forest cover, understory cover, elevation, planting density, pollination labor, vine length, and plant age as fixed effects, site as random effect, predictors scaled. Model selection used likelihood ratio tests (single-term deletions), retaining variables with p<0.05.

Vanilla yield averaged 105 ± 100 kg/ha and varied widely. Yield determinants: Yield increased with planting density (estimate 2.901, SE 0.415, p<0.001) and vine length (estimate 2.650, SE 0.393, p<0.001), and marginally with pollination labor input (estimate 0.897, SE 0.469, p=0.069). Yield showed no relationship with canopy closure, slope, landscape forest cover, understory cover, soil PC1, elevation, or plant age, and did not differ between fallow- vs forest-derived agroforests. Biodiversity–yield relationships: Across all taxa combined (mean normalized richness) and for trees, herbaceous plants, birds, amphibians, and ants, species richness (overall and endemic) generally did not relate to vanilla yield. Notable exceptions were a negative relationship between yield and butterfly species richness (estimate −0.179, p<0.001) and endemic butterfly richness (estimate −0.104, p=0.043), a positive relationship with amphibian richness (estimate 0.110, p=0.045), and land-use-history-dependent relationships for endemic reptiles (positive in forest-derived: estimate 0.289, p=0.028; negative in fallow-derived: estimate −0.145, p=0.016). Land-use history effects on plot-level richness: Forest-derived vanilla agroforests vs old-growth forest showed significant losses in birds (−38%), trees (−51%), amphibians (−51%), and mean normalized richness (−23%), while butterflies increased (+82%). Endemic richness declined strongly in forest-derived vanilla vs old-growth for amphibians (−57%), trees (−58%), birds (−69%), and overall endemic richness (−47%). Compared to forest fragments, only butterflies increased in forest-derived vanilla (+122%); other taxa showed no significant differences in overall or endemic richness. Fallow-derived vanilla vs fallows showed significant gains in overall tree richness (+149%), and in endemic richness for reptiles (+38%), ants (+164%), and mean normalized endemic richness (+38%). Gamma diversity was highest in old-growth forest for all taxa except some cases (overall herbaceous plants, butterflies, ants; endemic butterflies), and higher in fallow-derived vanilla than in fallows for most taxa. Endemism proportions among recorded species were high: trees 51%, herbaceous plants 20%, birds 61%, amphibians 98%, reptiles 74%, butterflies 58%, ants 45%. Species composition: Significant compositional shifts occurred from old-growth forest to forest-derived vanilla across taxa; forest fragments differed from forest-derived vanilla for several taxa; fallow-derived vanilla differed from fallows for trees, herbaceous plants, birds, reptiles, butterflies, and ants, but not amphibians. Management/environment effects on biodiversity: Yield-related variables showed trade-offs with tree and reptile diversity—higher planting density reduced tree richness (estimate −0.155, SE 0.066, p=0.019) and endemic tree richness (estimate −0.337, SE 0.096, p<0.001) but increased endemic herbaceous richness (estimate 0.176, SE 0.063, p=0.005). Longer vine length related to fewer tree species (estimate −0.221, SE 0.056, p=0.001), fewer endemic trees (estimate −0.553, SE 0.104, p=0.001), fewer reptiles (estimate −0.0166, SE 0.071, p=0.019), and marginally fewer endemic reptiles (estimate −0.096, SE 0.052, p=0.068). Non-yield variables were key biodiversity drivers: Canopy closure positively associated with trees (estimate 0.284, SE 0.082, p<0.001), reptiles (0.178, SE 0.074, p=0.016), endemic reptiles (0.136, SE 0.057, p=0.017), endemic ants (0.387, SE 0.111, p<0.010), and marginally with endemic herbaceous plants (0.145, SE 0.078, p=0.063). Understory cover increased endemic birds (0.460, SE 0.179, p=0.010) and tended to reduce butterflies (estimate −0.127, SE 0.072, p=0.077) and endemic butterflies (estimate −0.109, SE 0.049, p=0.062). Landscape forest cover increased tree (0.270, SE 0.074, p<0.001), endemic tree (0.827, SE 0.107, p<0.001), endemic herbaceous (0.236, SE 0.064, p<0.001), and endemic ant richness (0.214, SE 0.084, p=0.011), and marginally reduced amphibians (−0.097, SE 0.052, p=0.062). Amphibians declined with steeper slopes (−0.125, SE 0.054, p=0.021); herbaceous richness increased with elevation (0.283, SE 0.055, p<0.001).

The study demonstrates that within the observed management range, higher vanilla yields can be achieved without general losses in species richness for most taxa. This challenges assumptions of inevitable biodiversity–yield trade-offs in tropical smallholder agroforestry and highlights management levers that allow productivity gains alongside biodiversity conservation. Key to this is that vanilla yield was driven by planting density and vine length rather than by canopy opening or vegetation removal, while biodiversity was chiefly shaped by non-yield variables such as canopy closure and landscape forest cover. Consequently, farmers can intensify via more/longer vines while maintaining dense canopy structures and conserving nearby forest fragments to support multi-taxa biodiversity. Land-use history strongly determined biodiversity outcomes: forest-derived agroforests had substantially lower overall and endemic richness than old-growth forests, whereas fallow-derived agroforests recovered biodiversity relative to fallows, particularly for endemic taxa. These patterns affirm the irreplaceable conservation value of old-growth forests and the restoration potential of establishing vanilla agroforests on degraded/fallow land. Trade-offs emerged for trees and reptiles with higher planting density and vine length, indicating the need to manage these variables at intermediate levels to balance yields with structural habitat complexity provided by trees, which benefits multiple taxa. The negative association between yield and butterfly richness, and the mixed responses of endemic reptiles by land-use history, suggest taxa-specific sensitivities that may relate to understory density, microclimate, and host plant dynamics. Policy and practice implications include incentivizing fallow-derived agroforestry, maintaining canopy cover and forest fragments, and promoting standards or payment schemes that reward biodiversity-friendly practices without compromising yields.

Vanilla agroforestry in northeastern Madagascar can deliver win-win outcomes: high yields are generally compatible with high multi-taxa biodiversity when canopy structure is retained and forest remnants are conserved. Fallow-derived vanilla agroforests substantially enhance biodiversity compared to fallows and can help break cycles of land degradation, while old-growth forests remain irreplaceable for endemic-rich assemblages and require strict protection. Yield improvements hinge on planting density and vine length and need not involve reductions in canopy closure. Management strategies that balance intermediate intensification with maintenance of canopy and landscape forest cover can optimize both production and conservation. Future research should elucidate mechanisms behind observed trade-offs (e.g., vine length effects on tree and reptile communities), refine thresholds for planting density/vine length that minimize biodiversity costs, and assess long-term landscape-scale outcomes of incentivizing fallow-derived agroforestry.

Land-use history classification was binary and based on farmer interviews without detailed chronologies (e.g., burn frequency, prior crops). Sampling coverage averaged 84% but rarefaction curves for ants, butterflies, and herbaceous plants did not reach asymptotes in all land uses, indicating that additional sampling might amplify differences. Old-growth forest sampling was limited to two sites within one protected area. Yield was estimated via pod volume–weight regression rather than direct weighing at harvest, though theft and early harvest were ruled out in sampled pieds. Pollination labor input was imputed for three agroforests. Observational design limits causal inference, and some biodiversity–yield patterns may be mediated by unmeasured or collinear management/environmental variables.