Global systematic review with meta-analysis reveals yield advantage of legume-based rotations and its drivers

The study addresses how inclusion of legumes as pre-crops in cropping rotations affects the yield of subsequent main crops at a global scale and what factors drive the magnitude of this effect. While legume-inclusive rotations are proposed to enhance ecosystem services and reduce input dependence, their global yield impact and underlying drivers had not been systematically quantified. The authors hypothesize: (1) nitrogen (N) fertilization of the main crop reduces the legume pre-crop effect due to suppressed nodulation and N2 fixation; and (2) greater crop diversification in the initial non-legume system (in terms of species, functional groups, and temporal diversity) decreases the non-N break-crop benefits attributable to legumes. A comprehensive global synthesis is pursued to inform design and implementation of legume-based rotations across diverse pedo-climatic and management contexts.

Prior quantitative syntheses of legume pre-crop effects were limited either to specific crop species (e.g., cereals) or confined to regional scales. Reviews and meta-analyses have documented break-crop (non-N) effects and highlighted benefits of diversification, but lacked a unified global assessment accounting simultaneously for environmental and agronomic moderators such as initial crop diversification and N fertilization. The present work extends beyond earlier syntheses by including a broader set of legume purposes (grain, fodder, green manure) and main crop categories (cereals, oilseed, fiber, root, and tuber crops), aiming to disentangle N and non-N contributions across global conditions.

The authors conducted a PRISMA-guided systematic review across Web of Science, Google Scholar (first 5000 records), and CNKI up to October 2020 using rotation- and yield-related search terms. Inclusion criteria required field experiments with side-by-side comparisons of legume and non-legume pre-crop rotations leading to the same main crop, reported or inferable subsequent crop yield, consistent initial climatic/soil/management conditions, and stated location. Multiple experiments within publications were treated as distinct studies; repeated publications of the same experiment were merged; rotation cycles could be included independently; duplicated data were recorded once. The final dataset comprises 462 studies reported in 476 papers (1959–2020), covering 11,768 paired observations across 53 countries and 60 major legume-based cropping systems. Most subsequent crops were cereals (91%), with additional rapeseed (6%), roots and tubers (2%), and minor shares of fodder, fiber, and sugar crops. Approximately 45% of observations came from monocultures and 47% from short rotations (crop diversity ≤ 4) in the initial non-legume system. Data extraction included means, replication counts, and precision indicators when available; yields were standardized to kg ha−1 at crop-specific moisture contents. When only figures were provided, data were digitized; when only percentage changes were reported, reference yields were assumed for ln response ratio calculation. Site geolocation was recorded and used to extract climate (MAT, MAP, aridity index) from WorldClim and soil properties (pH, SOC, total N, texture) from HWSD when not reported. Management variables recorded included residue management, conservation tillage, irrigation, N fertilizer rate on the main crop, rotation cycle, and legume purpose. Crop diversity of the initial non-legume system was quantified as the product of number of crop species, number of crop functional groups, and number of crop species per year, capturing species, functional, and temporal diversity. Effect size was computed as the natural log response ratio lnRR = ln(X_L/X_C), where X_L and X_C are subsequent crop yields following legume and non-legume pre-crops, respectively. Due to limited availability of SD/SE (<30% of studies), observations were weighted by replication using W_i = (N_L×N_C)/(N_L+N_C). Observations with zero yields in treatment or control were excluded. To identify key moderators, a MetaForest (random forest for meta-analysis) approach was applied with 21 candidate moderators, incorporating observation weights, non-linearities, and interactions. Variable pre-selection and model tuning used recursive algorithms and 10-fold cross-validation to minimize RMSE. The optimized models (with 18 predictors after excluding most-missing variables) achieved R^2_oob ≈ 0.46–0.51 and R^2_cv ≈ 0.46–0.51. Variable importance consistently identified N fertilizer rate on the main crop and crop diversity of the initial system as top predictors, with low importance for residue management and irrigation. Subsequent meta-regression modeled lnRR as a function of N fertilizer rate (N), crop diversity (D), and their interaction using linear mixed-effects models (restricted maximum likelihood) with study as a random effect: lnRR = β0 + β1 N + β2 D + β3 N×D + β_study + ε. Residual normality violations prompted bootstrapping of coefficients (1000 iterations). Publication bias was assessed via Fail-Safe N (Rosenberg’s N_fs = 2,304,804, greatly exceeding the robustness threshold) and funnel plot symmetry, indicating no significant bias. Effect sizes were back-transformed to percentage change: [exp(lnRR) − 1] × 100. Analyses were performed in R 4.0.3. The dataset and code are publicly available in Figshare.

- Global effect: Across 11,768 paired observations from 462 field experiments in 53 countries (1959–2020), legumes increased subsequent main crop yield by 20.4% on average (median 10.2%; 95% CI: 17.7%–23.1%; P < 0.001). The distribution of effects was predominantly positive: 73.6% positive, 0.7% neutral, and 25.7% negative, with 50% of observations showing >10% increase. - Moderators: MetaForest and mixed-effects modeling identified N fertilizer rate on the main crop and crop diversity of the initial non-legume system as the dominant drivers of the legume pre-crop effect, outweighing climatic and edaphic factors. - Nitrogen fertilization: The yield advantage declined by approximately 7% for each additional 50 kg N ha−1 applied to the main crop. The negative slope of effect vs. N rate was weaker under higher crop diversity, indicating reduced sensitivity to N inputs in more diverse sequences. - Crop diversity: Each 1-unit increase in crop diversity (species × functional groups × crops per year) reduced the legume yield advantage by about 3.5% on average. Nevertheless, legumes retained some advantage even in relatively diverse systems. - Yield level dependence: Yield benefits were greatest at low initial yield levels and diminished sharply as initial yields increased, becoming negligible above the average yield level in the dataset. 79% of large responses (>20%) occurred below the average initial yield. - Regional variation: Average yield gains after legumes were 43% in Africa, 19% in North America, and 12% in Asia, underscoring stronger benefits in low-input regions. - Crop- and sequence-specific effects: The legume–cereal sequence showed a mean 21% advantage (CI: 18.8%–23.7%). Among 60 major cropping sequences, effects ranged from 2% (pea–rapeseed; CI: −2.9% to 7.7%) to 78% (mucuna–maize; CI: 59.1%–99.4%). By species, pigeon pea increased subsequent yields by 32.4% (CI: 23.8%–41.0%), versus 14.5% for common bean (CI: 9.5%–19.5%). Following crop responses included maize (+28.9%; CI: 25.5%–32.2%), rapeseed (+3.9%; CI: 0.0%–7.8%), cotton (−1.3%; CI: −12.0% to 9.5%), and buckwheat (+5.4%; CI: −3.5% to 14.2%). - Mechanistic interpretation: The declining benefit with higher N rates suggests that N provided by legumes (via biological N fixation and carryover) is a primary driver and can be partially substituted by mineral N. Increased crop diversity in non-legume systems confers break-crop (non-N) benefits that narrow the yield gap with legume-inclusive rotations.

The findings confirm that including legumes in rotations generally enhances subsequent main crop yields globally and clarify that the magnitude of this benefit is largely governed by N fertilization and initial crop diversity. The negative relationship with N inputs supports the hypothesis that legume-derived N is a central mechanism; higher mineral N supply diminishes the relative advantage of legume pre-crops by satisfying crop N demand and suppressing N2 fixation benefits. The negative association with crop diversity indicates that non-legume diversified rotations capture many break-crop benefits (reduced biotic and abiotic stress through improved soil properties), thereby reducing the incremental advantage of legumes. Nonetheless, legumes maintain a positive effect even in diverse systems, aligning with concerns that reduced legume use in rotations may contribute to yield stagnation in intensive regions. Critically, the strong benefits observed in low-yield, low-input contexts (e.g., parts of Africa and in organic systems) demonstrate the potential of legume integration to improve productivity where inputs are constrained, while in high-yield systems legume inclusion offers opportunities to reduce N inputs with minimal yield penalty. Sequence- and species-specific variability highlights the importance of locally optimized rotation design, considering legume type, main crop, and management practices (e.g., conservation tillage, residue management).

This global meta-analysis demonstrates that legume-based rotations increase subsequent main crop yields by about 20% on average, with especially strong impacts in low-input, low-yielding systems. Across climates and soils, N fertilization and initial crop diversity are the principal moderators, with higher N rates and greater diversity reducing the legume pre-crop advantage. The results support legume inclusion as a key pathway for sustainable intensification—enhancing yields, enabling reduced N fertilizer use, and contributing to diet quality via plant protein—while acknowledging potential short-term trade-offs in main crop area. Future work should refine rotation designs to maximize benefits in specific sequences and environments, integrate complementary practices such as conservation tillage and residue retention, and further disentangle N versus non-N mechanisms to guide N management and diversification strategies.

- Data heterogeneity and reporting: Standard deviations/standard errors were available for fewer than 30% of studies, necessitating replication-based weighting rather than inverse-variance weighting, which may affect precision of pooled estimates. Some data were digitized from figures or inferred from percentage changes, introducing potential extraction error. - Missing covariates: Climate and soil variables were imputed from global databases when not reported; in a small number of cases with only broad locations, environmental variables could not be estimated, potentially adding noise to moderator analyses. - Coverage imbalance: The dataset is dominated by cereal main crops and certain regions, which may limit generalizability to underrepresented crops or areas. - Model assumptions: Some mixed-effects models violated normality; coefficients were bootstrapped to mitigate this, but residual heterogeneity remains. - Practical trade-offs: While yield benefits are observed, legume cultivation can reduce main crop area in the short term, affecting immediate production metrics; benefits depend on context, N input levels, and rotation diversity.