How rice land availability shapes diversification–specialisation gradients in Madagascar's clove landscapes

doi:10.21203/rs.3.rs-8395888/v1

Download PDF

Research Article

How rice land availability shapes diversification–specialisation gradients in Madagascar's clove landscapes

https://doi.org/10.21203/rs.3.rs-8395888/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Understanding why agroforestry systems are more or less diversified in species is essential for sustaining livelihoods and ecological functions in perennial-crop landscapes. In eastern Madagascar, clove-based agroforestry dominates upland production systems, yet field structure varies markedly across short distances. Our objective was to clarify how landscape-scale rice production capacity influences agroforestry diversification–specialisation trajectories in clove-dominated landscapes. We combined tree-inventory data from 54 plots with landscape descriptors and a quantitative proxy of village-level rice self-sufficiency to analyse agroforestry composition across three neighbouring villages with contrasting access to irrigated lowlands. Four agroforestry system types—complex agroforests, diversified cash-oriented systems, clove-dominated systems and simple parklands—were identified through rule-based classification. Village-level access to irrigated lowlands was strongly associated with their distribution. In the valley with extensive irrigated lowlands, agroforestry systems remained highly diversified, integrating fruit, forest and cash-crop species. In intermediate landscapes, mixed configurations occurred. In upland areas where irrigated lowlands were scarce, clove became the main means of securing staple foods through market purchases, corresponding to simplified, clove-dominated systems. Contrary to expectations, within-village topography did not influence perennial species composition. Village-scale differentiation aligned with food-security constraints operating at the household level rather than fine-scale ecological sorting. These findings suggest that contemporary agroforestry configurations result from the interaction between historical trajectories—colonial coffee plantations, farmer-led clove expansion and post-independence enrichment practices—and present-day rice availability, which appears to filter how diversification or specialisation unfolds. Our results highlight that maintaining agroforestry diversity in clove-producing regions depends as much on securing stable rice production systems as on managing perennial-crop markets.

Agroforestry systems

Rice self-sufficiency

Clove production

Landscape gradients

Species diversity

Eastern Madagascar

Agroforestry systems are central to tropical smallholder agriculture, most often combining perennial cash crops, food production, and multifunctional tree cover that supports biodiversity and resilience of farmers (Schroth et al., 2004; Tscharntke et al., 2011; Beillouin et al., 2021; Terasaki Hart et al., 2023). Tree species diversity and vertical stand structure (canopy stratification, density, and spatial arrangement) emerge from the interaction of three main sets of drivers: environmental characteristics (climate, soils, topography), farmers’ strategies (labour allocation, risk management, food security and income objectives) and historical and socioeconomic context (colonial heritage, dynamics of cash crop prices for export and local markets, demographic pressure and tenure changes) (Jose, 2009; Tittonell, 2014). Understanding how these multiscale and multitemporal interactions determine agrobiodiversity in agroforestry systems within and between landscapes is essential for identifying pathways toward sustainable and resilient smallholder systems.

Along Madagascar's eastern coast, agroforestry mosaics dominated by clove trees, vanilla plants, coffee, lychee and cinnamon trees are embedded within a marked contrast between irrigated lowlands (horaka) and steep upland hillslopes (tanety) (Danthu et al., 2024). These landscapes are the product of successive socio-historical transformations, each leaving distinct imprints on contemporary land use. Precolonial settlement (pre-1897) was organised around irrigated rice in valley bottoms (horaka), with shifting cultivation (tavy) generating a dynamic matrix of fallows and secondary vegetation on hillslopes (Blanc-Pamard and Ruf, 1992). Colonial administration (1897–1960) profoundly restructured uplands through the successive imposition of, first, linear coffee plantations under Albizia shade, followed by clove tree plantations, established through head-tax incentives and coercive labor regimes (Fremigacci, 1986; Danthu et al., 2022). Following mid-20th-century coffee crises and the progressive abandonment of coffee as an export crop, clove trees expanded across upland areas through the conversion of both coffee plantations and tavy fallows (savoka).

Since Madagascar's independence, farmers have progressively enriched these clove systems with fruit, timber and forest species to mitigate the impact of price volatility for clove products on international markets (Arimalala et al., 2019; Michel et al., 2021). More recently, farmers have introduced lychee trees and vanilla vines cultivated on living supports into their upland systems, with production destined for export markets. This historical succession—from tavy to clove—underpins the highly heterogeneous agroforestry landscapes observed today. In these landscapes, simplified clove-dominated systems with sparse associated species coexist with complex, multistrata agroforests that integrate fruit, timber and forest trees.

Today, the clove tree is the agricultural and economic pillar of the eastern coast, positioning Madagascar among the world's leading exporters of clove dried buds and essential oil (Danthu et al., 2024). Yet the diversity of clove-based system structures and species compositions reflect more than ecological suitability. Tree stand characteristics—including clove tree density, canopy stratification, spatial arrangement of associated trees, and woody species diversity—result from how individual households and communities navigate labour constraints, land scarcity, demographic pressure, and exposure to climatic and economic risks. The Analanjirofo region, Madagascar's main clove-producing area, is marked by chronic vulnerability shaped by recurrent cyclones, high market volatility, structural isolation, and persistent poverty despite its favourable agro-ecological conditions (Droy et al., 2017). These interacting risks strongly influence livelihood choices: cyclones damage tree plantations and reduce rice harvests, while clove price fluctuations affect household cash flow, influencing whether farmers maintain diversified tree stands or specialise in clove production.

A central dimension of these trade-offs is rice production, the primary staple food and main marker of food security and social stability in eastern Madagascar (Dabat et al., 2008; Droy et al., 2017). Access to irrigated lowlands determines whether households can meet their own rice needs or must rely on market purchases funded through cash-crop revenues. When rice production is sufficient, agroforestry plots can serve multiple functions beyond cash income. Conversely, when irrigated lowlands are scarce, households must rely on cash-crop revenues to purchase rice (Andriatsitohaina et al., 2020). However, how this dependence on cash crop revenues to meet food needs influences the species composition, tree stand structure, and the diversification–specialisation trajectories of clove-based systems remains unknown.

Most existing studies focus either on individual farmer management of agrobiodiversity at the plot scale or on broad landscape-scale dynamics, but rarely integrate food-system constraints into the analysis of tree species composition and stand structure in tree-based cropping systems. In eastern Madagascar, several contributions clarify key mechanisms shaping clove systems diversity: topography and soil conditions structure species assemblages and system configurations (Mariel et al., 2022a); long-term transitions from tavy to coffee and then clove have reshaped landscape mosaics and species pools (Mariel et al., 2021a); and knowledge transmission and social networks influence the diffusion of tree species (Mariel et al., 2022b). Yet despite these insights, the role of irrigated rice paddy availability and the capacity to ensure rice food supply at the village scale in driving diversification–specialisation patterns within clove-based systems remains unexamined.

The Iazafo landscape (Analanjirofo region) provides an ideal setting to analyse these dynamics (Tonneau et al., 2018). Over a few kilometers, it spans a steep gradient in irrigated lowland availability, slope steepness, settlement density and agrarian history—from the wide rice plains of Ampahibe to the uplands of Andampy. These contrasts create strong variations in rice self-sufficiency capacity and livelihood strategies, offering a natural experiment to analyse how landscape context shapes tree density and species richness in clove-based systems. This gradient-based setting allows us to test whether the capacity for rice autonomy through irrigated paddy availability acts as a primary structuring force or whether other factors—particularly topographic heterogeneity—exert stronger local effects on tree density and species richness in clove-based systems.

Our objective is to clarify how landscape-scale rice production capacity at the village territory scale influences agroforestry diversification–specialisation trajectories in clove-dominated landscapes. To address this objective, we hypothesised that variation in village-level access to irrigated rice lowlands generates systematic differences in the structure, composition and species diversity of clove systems, whereas fine-scale topography exerts a more limited influence once systems are established.

We tested this hypothesis through comparative analysis of tree stand structure and species composition in clove-based systems distributed along transects across three contrasting villages. Clove-based systems were characterised using floristic inventories, structural indicators (tree density, basal area, diameter at breast height), and functional species groups, combined with multivariate analyses and a rule-based typology aligned with regional agroforestry literature. This approach enabled us to identify distinct system types and assess how rice paddy availability shapes their spatial distribution and agrobiodiversity patterns.

Study area

The study was conducted in the Analanjirofo region, in eastern Madagascar, a humid tropical region receiving > 2,000 mm of rainfall annually and mean temperatures near 25°C. This work focused on the Iazafo landscape, a mosaic of irrigated rice lowlands (horaka) embedded within cultivated hillslopes (tanety) between Vavatenina and Fenoarivo Atsinanana, and traversed by the RN22 national road (Tonneau et al., 2018). Elevation rises from around 100 m near the coast to > 1,500 m inland (Dandoy, 1973).

Three villages (fokontany)—Ampahibe, Antsirakoraka and Andampy—were selected along a pronounced gradient in irrigated lowland extent, slope configuration and land availability (Fig. 1; Table 1). Lowland delineation followed landscape diagnosis of the PADAP project (Tonneau et al., 2018). Ampahibe (Fig. 2A) lies within wide irrigated plains with gentle slopes. Antsirakoraka (Fig. 2B) consists of alternating narrow valley bottoms and short slopes. Andampy (Fig. 2C) is a steep hillside landscape with almost no irrigated rice land. Soils consist of deeply weathered ferralitic substrates differentiated by slope position (Dandoy, 1973; Mariel, 2022). Population densities range from 200 to 400 inhabitants per km², however steep slopes restrict cultivable land to approximately 0.6 ha on average per household (Tonneau et al., 2018).

Table 1

Characteristics of the three study villages and rice self-sufficiency estimates based on irrigated paddy availability at the village scale
Village	Commune (District)	Distance to RN22 (km)	Area (km²)	Irrigated lowlands (ha)ᵃ	Population (inhab.)ᵇ	Density (inhab·km⁻²)	Milled rice production (kg·person⁻¹·year⁻¹)ᶜ	Rice self-sufficiency (%)ᵈ	Landscape structure and dominant land use
Ampahibe	Ampasimazava (Vavatenina)	≈ 9	4.2	176	1,675	399	176	+ 15	Flat valley landscape forming the core of the Iazafo plain, dominated by wide irrigated rice lowlands. Surrounding gentle slopes support trees and annual crops but with sparse and fragmented tree cover. Dense settlement and road networks.
Antsirakoraka	Maromitety (Vavatenina)	≈ 2.5	5.5	69	1,087	198	106	-30	Mosaic of alternating slopes and narrow irrigated lowlands. Rice persists in valley bottoms; upper slopes host clove-based agroforests mixed with coffee, vanilla, and annual crops. Tree cover is heterogeneous, forming mid-density agroforestry mosaics along short topographic sequences.
Andampy	Mahambo (Fenoarivo Atsinanana)	≈ 7	5.5	15	1,401	255	18	-90	Steep hillside landscape with continuous cultivated slopes and almost no irrigated lowlands. Clove-dominated agroforestry systems cover most hillsides; few annual crops occur. Landscape is fully hillside-based, with agroforests occupying nearly all cultivable surfaces.

ᵃIrrigated lowland area derived from QGIS spatial analyses (this study)

ᵇPopulation from local census: Antsirakoraka 2023 (Be and Sarron, 2023); Ampahibe and Andampy 2024 (Ramalanjaona, 2025)

ᶜMilled rice production calculated as: Irrigated area × mean seasonal paddy yield (2.5 t ha⁻¹; Stratégie Nationale de Développement Rizicole III: 2022–2030, 2022) × milling recovery (67%) ÷ Village population

ᵈRice self-sufficiency = [(Milled rice per capita ÷ National consumption benchmark) − 1] × 100, where national per-capita consumption = 153 kg person⁻¹ yr⁻¹ (International Monetary Fund. African Dept., 2025). Positive values indicate surplus; negative values indicate deficit relative to national consumption levels.

Sampling design and field surveys

Fieldwork was conducted between March and June 2025. To capture floristic and structural variation across the landscape gradient, a stratified sampling was designed based on transects across the hillslopes.

Hillslope transects

In each village, six transects (18 transects in total) were positioned along continuous toposequences extending from the valley bottom—or its immediate border—to the first local summit (Fig. 3). Transects were established on both east- and west-facing slopes. In Ampahibe, east-facing slopes were entirely occupied by irrigated fields, so all transects were west-facing. Minimum and maximum elevations were recorded using a Garmin GPSMAP 64s. Elevation ranges within each transect were divided into three equal intervals, defining lower-, mid-, and upper-hill sections.

Vegetation plots and woody-inventory protocol

A 20 × 20 m (400 m²) plot was established in each hill section (lower-, mid- and upper-hill) of each transect, yielding 54 plots in total, all located in areas of continuous tree cover (Table 2; Fig. 3). Within each plot, all woody individuals (trees, shrubs and palms) with a circumference at breast height (CBH) ≥ 5 cm were inventoried. For each individual, we recorded: (i) vernacular and scientific names; (ii) the number of stems; and (iii) CBH for each stem using a measuring tape, and total height using a Vertex 5 hypsometer. Young clove trees < 5 cm CBH were excluded from structural analyses.

For each individual, basal area (cm²) was calculated by summing the basal areas of all stems. For multi-stemmed individuals, an equivalent circumference was then derived from the total basal area to obtain a realistic single-stem equivalent value. All stem counts and basal area values were converted to per-hectare values (stems ha⁻¹ and m² ha⁻¹) based on the 400 m² plot area.

Table 2

Altitudinal ranges (m a.s.l.) of lower-, mid- and upper-slope plots in the three study fokontany
Village	Lower-hill (m)	Mid-hill (m)	Upper-hill (m)
Ampahibe	124–147	141–170	182–204
Antsirakoraka	116–156	135–270	155–300
Andampy	137–221	148–256	171–311

Understorey and associated crops

Understorey cultivated species were systematically recorded. Stems of vanilla (Vanilla planifolia), banana (Musa spp.), and pineapple (Ananas comosus) were counted. Other perennial crops grown under partial shade (e.g., Manihot esculenta, Saccharum officinarum) were noted as present/absent. For annual crops (e.g., Oryza sativa, Zea mays), we used a binary variable indicating whether rice or maize was actively cultivated within the plot during the survey. This variable was used to characterise crop–tree integration within agroforestry systems.

Processing of floristic and structural data

Groups of uses

All woody species were assigned to one of three use groups. Because vanilla (V. planifolia) is a non-woody vine, its presence was represented through the woody support trees on which it grows. Cash-crop species included perennial species primarily cultivated for sale. This group comprised the main export-oriented cash-crop species—clove trees (Syzygium aromaticum), vanilla lianes (V. planifolia, represented via support trees) and lychee trees (Litchi chinensis)—as well as coffee trees (Coffea canephora), which was historically a cash crop but is now predominantly traded on local markets. Minor cash-crop tree species (such as Cinnamomum sp. or Elaeis guineensis) were retained in the full species inventory but did not structure use grouping. Fruit species corresponded to tree crops grown mainly for household consumption or for local market channels, such as Mangifera indica, Persea americana, Citrus spp., Artocarpus altilis, Psidium guajava and Cocos nucifera.

Forest tree species encompassed multipurpose woody species primarily used for timber, poles, fuelwood, or for ecological and agronomic functions such as shade provision, windbreaks, and soil improvement within agroforests. This group included, among others, Albizia spp., Grevillea banksii, Harungana madagascariensis, Eucalyptus spp., Bambusa spp. and Raphia farinifera.

Use group assignments were validated with farmers during fieldwork.

System-structure metrics

For each plot, we quantified total stem density (stems ha⁻¹), total basal area (m² ha⁻¹), stem density and basal area for each use group, and clove dominance, expressed as the stem density and basal area of S. aromaticum. Together, these indicators extend the structural framework proposed by Michel et al. (2021) and allow us to position systems along a diversification–specialisation continuum.

To characterise system structural development independently of species identity, we then classified plots into three structural stages based on the empirical distribution of total basal area. Total basal area (m². ha⁻¹) was calculated for each plot, and the 35th and 80th percentiles of its landscape-wide distribution were used as data-driven thresholds. Plots with basal area ≤ P₃₅ were classified as open, those ≥ P₈₀ as dense, and those remaining as intermediate. These structural stages provide a simple quantitative index of canopy development that integrates both stem number and stem size.

Alpha diversity and dominance

Alpha-diversity was quantified at the plot level to compare species composition and dominance patterns among systems of similar spatial extent. For each plot, we calculated species richness (S), defined as the total number of species recorded; Shannon diversity (H'; Shannon, 1948) ; Pielou’s evenness (J; Pielou, 1966) ; and Berger–Parker dominance (d; Berger and Parker, 1970). We additionally calculated the inverse Simpson index (1/D) to express diversity in terms of effective numbers of species (Simpson, 1949). All diversity indices were computed using relative stem densities, allowing consistent comparisons across plots differing in structure or stem abundance.

Typology of clove cropping system

We developed a rule-based typology to characterise clove systems, drawing on the structural descriptors initially proposed for clove-based systems by Michel et al. (2021) and on the decision-based classification logic used by Mariel et al. (2021b) in the same clove-growing landscapes of eastern Madagascar. The typology combines three complementary dimensions. First, overall structural development was described using three basal-area classes (open, intermediate, dense) derived from the distribution of total basal area of trees. These classes provide a simple, quantitative index of canopy development independent of species identity. Second, clove tree dominance was assessed from clove stem density and from the contribution of S. aromaticum to total basal area. Third, the use-based species composition of each system was quantified from the relative structural contribution of cash-crop, fruit and forest species. Thresholds combining these three dimensions were then used to assign each plot to a system category following transparent, reproducible rules rather than multivariate clustering.

Statistical analyses

Analyses were performed in R (version 4.2.3). Pairwise comparisons of diversity and structural variables among cropping-system types were conducted using Kruskal–Wallis tests followed by Dunn post-hoc tests with Benjamini–Hochberg (BH) correction (FSA package; Ogle et al., 2023). Fisher's exact tests were used to assess whether village identity affected the distribution of system types.

To account for the nested sampling design (plots within transects within villages), Shannon diversity was analysed using a linear mixed-effects model with village as a fixed effect and transect nested within village as a random effect. Mixed models were fitted with lme4 and lmerTest packages for inference (Bates et al., 2015; Kuznetsova et al., 2017).

All compositional analyses were conducted using the vegan R package (Oksanen et al., 2022). Floristic dissimilarity among plots was quantified using Bray–Curtis distances (Bray and Curtis, 1957) computed from species-level stem densities. Bray–Curtis provides a measure of compositional difference that is sensitive to changes in both species identity and relative abundance. Non-metric multidimensional scaling (NMDS; Kruskal, 1964) was used to visualise gradients in woody species composition among plots. Differences in multivariate composition among villages were tested using permutational analysis of variance (PERMANOVA; Anderson, 2001, adonis function), with permutations constrained within transects to respect the hierarchical sampling design; both constrained and unconstrained models yielded consistent results. Homogeneity of multivariate dispersion—a key assumption for interpreting PERMANOVA—was assessed using PERMDISP (Anderson, 2001, betadisper function). NMDS ordinations were produced with metaMDS, and similarity percentage (SIMPER ; Clarke, 1993, SIMPER function) was used to identify the species contributing most to between-village dissimilarities.

Four clove cropping systems along a diversification–specialisation gradient

Applying the rule-based classification that combines basal area class, clove dominance and use-based composition, the 54 plots were grouped into four system types that span a clear gradient from simplified, clove-specialised systems to diversified agroforests (Table 3).

Simple parkland

Simple parklands (n = 6) correspond to open systems with annual crops in the understory. Total basal area was the lowest among all types (3.6 ± 2.2 m² ha⁻¹). Cash-crop basal area represented 44.2 ± 37.9% of total basal area (1.1 ± 1.4 m² ha⁻¹). In terms of stem density, clove stems contributed 40.1 ± 39.7%, forest stems 47.6 ± 35.9%, while fruit trees were sparse (11.9 ± 8.1% of stems). Species richness and biodiversity were moderate (6.2 ± 3.5 species; H' = 1.19 ± 0.69), reflecting a relatively simple tree layer over cropland rather than a fully developed agroforest (Table 3).

Clove-dominated system

Clove-dominated systems (n = 10) occurred mainly in intermediate or dense basal area classes, with a total basal area of 14.4 ± 8.4 m² ha⁻¹. The cash-crop basal area represented 78.8 ± 21.8% of total basal area, almost entirely composed of clove (clove basal area 11.1 ± 8.3 m² ha⁻¹). S. aromaticum accounted for more than 80% of stems in most plots (mean 87.3 ± 8.8%), with high clove density (566.1 ± 194.4 stems ha⁻¹). Species richness and diversity were the lowest of all types (3.0 ± 1.4 species; H' = 0.44 ± 0.29; Table 3). These systems thus correspond to simplified clove plantations, structurally well-developed but with very limited diversification beyond the main cash crop.

Diversified cash-oriented system

Diversified cash-oriented systems (n = 8) corresponded to intermediate basal area class systems, with a total basal area of 9.7 ± 8.6 m² ha⁻¹. These systems combined clove with at least one additional cash-crop species (2.1 ± 0.4 cash-crop species on average, including vanilla, coffee, or lychee). The cash-crop basal area represented 83.4 ± 21.7% of total basal area, with clove contributing roughly two-thirds (clove basal area 6.2 ± 7.7 m² ha⁻¹). In terms of stem density, cash-crop stems represented 83.1 ± 9.5% of total stems, while fruit and forest species remained structurally secondary (fruit stems 5.1 ± 7.1%, forest stems 11.9 ± 6.1%). Species richness and diversity were intermediate between clove-dominated and complex agroforests (5.2 ± 1.7 species; H' = 0.89 ± 0.35), confirming that these systems are strongly commercial but partially diversified within the cash-crop layer (Table 3).

Complex agroforests

Complex agroforests (n = 30) were predominantly dense systems, with the highest total basal area on average (18.7 ± 17.8 m² ha⁻¹). The cash-crop basal area represented only a minor fraction of total basal area (16.5 ± 20.3%), while fruit and forest basal area together dominated the structure (fruit basal area 9.5 ± 9.8 m² ha⁻¹, forest basal area 6.2 ± 13.4 m² ha⁻¹). In terms of stem density, fruit and forest species accounted for around 80% of stems (fruit stems 33.2 ± 21.1%, forest stems 44.9 ± 25.2%), while clove was generally present but structurally minor (14.6 ± 13.0% of stems; 8.2 ± 15.3% of basal area). These systems were numerically the most frequent (30 out of 54 plots) and showed the highest species richness and diversity (8.1 ± 3.1 species; H' = 1.69 ± 0.44), with more fruit (3.4 ± 1.9) and forest species (3.2 ± 1.7) than other system types (Table 3).

Table 3

Structural and compositional contrasts among clove system types. Means ± SD followed by different letters indicate differences among system types (Dunn's post-hoc test with Bonferroni correction, p < 0.05).
Category	Variable	Diversified cash-oriented (8)	Clove-dominated (10)	Simple parkland (6)	Complex agroforest (30)	Kruskal–Wallis H
Overall diversity	Species richness (no. spp)	5.2 ± 1.7ᵃᵇ	3.0 ± 1.4ᵇ	6.2 ± 3.5ᵃᵇ	8.1 ± 3.1ᵃ	24.15***
	Shannon diversity index (H′)	0.9 ± 0.4ᵇ	0.4 ± 0.3ᵇ	1.2 ± 0.7ᵃᵇ	1.7 ± 0.4ᵃ	30.98***
	Total basal area (m² ha⁻¹)	9.7 ± 8.6ᵃᵇ	14.4 ± 8.4ᵃᵇ	3.6 ± 2.2ᵇ	18.7 ± 17.8ᵃ	11.61**
Cash crops (all)	No. of cash-crop species	2.1 ± 0.4ᵇ	1.1 ± 0.3ᵃ	1.0 ± 0.6ᵃ	1.5 ± 0.9ᵃᵇ	12.07**
	Cash-crop tree density (stem. ha⁻¹)	575.0 ± 237.9ᵇ	566.1 ± 194.4ᵇ	254.2 ± 221.0ᵃᵇ	129.2 ± 100.2ᵃ	30.46***
	Cash-crop basal area (m² ha⁻¹)	7.7 ± 7.7ᵃᵇ	11.6 ± 8.1ᵇ	1.1 ± 1.4ᵃ	3.0 ± 4.5ᵃ	20.12***
	Share of cash-crop stem density (%)	83.1 ± 9.5ᵇ	87.9 ± 9.1ᵇ	40.5 ± 39.8ᵃᵇ	21.9 ± 15.4ᵃ	31.65***
	Share of cash-crop basal area (%)	83.4 ± 21.7ᵇ	78.8 ± 21.8ᵇ	44.2 ± 37.9ᵃᵇ	16.5 ± 20.3ᵃ	30.11***
Clove (specific)	Clove tree density (stem. ha⁻¹)	453.1 ± 319.4ᵇ	563.6 ± 196.2ᵇ	250.0 ± 216.2ᵃᵇ	85.8 ± 82.4ᵃ	25.48***
	Clove basal area (m² ha⁻¹)	6.2 ± 7.7ᵇᶜ	11.1 ± 8.3ᵇ	1.1 ± 1.3ᵃᶜ	1.1 ± 2.0ᵃ	24.96***
	Share of clove stem density (%)	60.1 ± 31.0ᵇ	87.3 ± 8.8ᵇ	40.1 ± 39.7ᵃᵇ	14.6 ± 13.0ᵃ	28.19***
	Share of clove basal area (%)	65.5 ± 33.7ᵇ	76.3 ± 25.3ᵇ	43.2 ± 37.3ᵃᵇ	8.2 ± 15.3ᵃ	27.99***
	% of clove trees 15–20 cm DBH	18.2 ± 25.3ᵇᶜ	26.5 ± 14.6ᵇ	1.9 ± 4.5ᵃᶜ	7.6 ± 21.3ᵃ	24.11***
Fruit trees	No. of fruit species	0.9 ± 1.5ᵇ	0.9 ± 0.7ᵇ	2.0 ± 1.3ᵃᵇ	3.4 ± 1.9ᵃ	22.38***
	Fruit tree density (stem. ha⁻¹)	31.2 ± 43.8ᵇ	45.6 ± 40.6ᵇ	66.7 ± 40.8ᵃᵇ	187.5 ± 122.6ᵃ	20.83***
	Fruit basal area (m² ha⁻¹)	0.4 ± 0.7ᵇ	2.5 ± 4.1ᵇ	0.8 ± 0.6ᵃᵇ	9.5 ± 9.8ᵃ	22.11***
	Mean fruit-tree height (m)	1.5 ± 2.2ᵇ	5.9 ± 6.2ᵃᵇ	4.1 ± 3.1ᵃᵇ	8.6 ± 4.6ᵃ	18.10***
	Share of fruit stem density (%)	5.1 ± 7.1ᵇ	7.4 ± 6.5ᵇ	11.9 ± 8.1ᵃᵇ	33.2 ± 21.1ᵃ	23.81***
	Share of fruit basal area (%)	3.0 ± 4.3ᵇ	15.2 ± 20.2ᵇ	20.1 ± 14.8ᵃᵇ	50.7 ± 32.2ᵃ	23.54***
Forest trees	No. of forest species	2.2 ± 1.3ᵃᵇ	1.0 ± 1.3ᵇ	3.2 ± 2.5ᵃᵇ	3.2 ± 1.7ᵃ	13.15**
	Forest tree density (stem. ha⁻¹)	81.2 ± 54.7ᵃᵇ	28.1 ± 34.7ᵇ	308.3 ± 289.3ᵃ	287.5 ± 227.4ᵃ	21.39***
	Forest basal area (m² ha⁻¹)	1.6 ± 3.4ᵃᵇ	0.3 ± 0.7ᵇ	1.7 ± 1.8ᵃᵇ	6.2 ± 13.4ᵃ	16.00**
	Share of forest stem density (%)	11.9 ± 6.1ᵇᶜ	4.7 ± 5.6ᵇ	47.6 ± 35.9ᵃᶜ	44.9 ± 25.2ᵃ	23.69***
Annual crops	Presence of annual crops (0–1)	0.0 ± 0.0ᵃ	0.0 ± 0.0ᵃ	0.8 ± 0.4ᵇ	0.0 ± 0.0ᵃ	43.27***

Kruskal–Wallis global tests: *p < 0.05, **p < 0.01, ***p < 0.001.

Within-village variation of diversity along hill section

Alpha diversity along hill sections

Across the landscape, the Shannon index (H’) did not differ significantly among lower-, mid- and upper-hill plots (Kruskal–Wallis p = 0.11). The same absence of pattern was observed within each village, where none of the pairwise contrasts among hill sections remained significant after adjusting p-values with the Benjamini–Hochberg (BH) procedure (all BH-adjusted p ≥ 0.14). Diversity tended to be slightly lower on upper hills in Antsirakoraka and Andampy, but these differences were not statistically supported once BH correction was applied.

Together, these results indicate that fine-scale topography did not significantly influence plot-level woody diversity.

Composition of groups of use and key tree species along hills

Within each village, the structural contributions of the three use groups—cash-crop, fruit trees, and forest species—did not differ significantly among hill sections (all BH-adjusted p ≥ 0.45). Hill position had no detectable effect on the densities of the main tree crops—clove tree, vanilla and coffee—as none of the contrasts among hill sections were significant (all BH-adjusted p ≥ 0.36). Hill orientation had only a marginal influence on clove density. In Andampy, clove density did not differ between east- and west-facing hills (p = 0.83). In Antsirakoraka, west-facing hills tended to support slightly higher clove densities, but the contrast remained close to the significance threshold (BH-adjusted p = 0.051). Overall, hill position and hill orientation had no detectable influence on the composition of use groups or on the distribution of key tree crops.

Distribution of cropping-system types along hill positions

All four cropping-system types occurred across lower-, mid- and upper-hill sections in each village. Kruskal–Wallis tests confirmed the absence of significant differences in the distribution of system types across hill sections after BH correction (all adjusted p > 0.10). Thus, within-village spatial organisation of agroforestry types appears to be independent of hill position, confirming the limited role of micro-topography in shaping system structure.

Differences between villages

Floristic and structural diversity across villages

Across the 54 agroforestry plots, we recorded 45 woody species from 20 plant families. Clear and consistent contrasts emerged along the landscape gradient from Ampahibe (rice-rich valley bottom) to Antsirakoraka (short hill sequences) and Andampy (fully hillside landscape). Species richness declined sharply from Ampahibe (16.0 ± 4.9 species per plot) to Antsirakoraka (12.0 ± 4.7) and reached its lowest values in Andampy (7.0 ± 3.8). Shannon diversity followed the same pattern (Ampahibe = 1.68 ± 0.42; Antsirakoraka = 1.38 ± 0.65; Andampy = 0.82 ± 0.58), while Berger–Parker dominance increased strongly toward Andampy, indicating increasingly simplified and uneven assemblages. Tree density increased moderately from Ampahibe (535 ± 235 stems ha⁻¹) to Andampy (686 ± 191 stems ha⁻¹), suggesting that simplified systems are not less stocked but are structurally dominated by a few species.

Village-level differences in alpha-diversity were first detected using non-parametric tests (Kruskal–Wallis; Table 4). A linear mixed-effects model confirmed a strong village effect on Shannon diversity (F₂,₁₅ = 9.58, p = 0.002), and estimated marginal means closely matched plot-level patterns (Ampahibe 1.68 ± 0.14; Antsirakoraka 1.38 ± 0.20; Andampy 0.82 ± 0.20). This shows that the decline in alpha-diversity from Ampahibe to Andampy remains robust after accounting for the nested sampling design (plots within transects within villages).

Table 4

Floristic and structural diversity across the three fokontany of the Analanjirofo clove landscape (East Madagascar). Values are means calculated at plot level (n = 18 plots per fokontany). Species richness (S), Shannon diversity (H') and tree density are reported as mean ± SD; Berger–Parker dominance (d), effective diversity (1/D) and evenness (Pielou J) are reported as mean values. Shannon diversity, Berger–Parker dominance and evenness were computed from species-level relative densities per plot. Letters denote significant pairwise differences among fokontany (Wilcoxon rank-sum tests with Bonferroni correction, p < 0.05)
Village	Species richness^a	Shannon index (H’)^b	Tree density (stems. ha^− 1)	Berger-Parker dominance (d)	Effective diversity (1/D)	Evenness (J)
Ampahibe	16.0 ± 4.9a	1.68 ± 0.42a	535 ± 235	0.37	5.03	0.86
Antsirakoraka	12.0 ± 4.7a	1.38 ± 0.65a	657 ± 228	0.51	3.82	0.72
Andampy	7.0 ± 3.8b	0.82 ± 0.58b	686 ± 191	0.72	2.20	0.58
^aSpecies richness: Kruskal–Wallis χ² = 10.56, p = 0.005
^bShannon diversity: Kruskal–Wallis χ² = 16.10, p < 0.001; LMM confirmed a strong village effect (F₂,₁₅ = 9.58, p = 0.002)

PERMANOVA on Bray–Curtis dissimilarities detected significant differences in species composition among villages (R² = 0.153, F₂,₅₁ = 4.60, p = 0.001). Multivariate dispersion did not differ among villages (PERMDISP: F₂,₅₁ = 2.10, p = 0.13; mean distance to centroid ± SD: Ampahibe 0.50 ± 0.12, Antsirakoraka 0.52 ± 0.14, Andampy 0.40 ± 0.26), indicating that compositional contrasts are due to centroid separation rather than unequal within-village heterogeneity (Fig. 4). A hierarchical PERMANOVA with permutations constrained within transects produced identical effect sizes and significance (R² = 0.153, F₂,₅₁ = 4.60, p = 0.001), confirming that village-level differences are robust to the nested sampling structure.

NMDS ordination based on Bray–Curtis distances revealed a clear separation between Ampahibe and Andampy, with Antsirakoraka occupying an intermediate position in ordination space. SIMPER analysis showed that the clove tree was the main contributor to between-village dissimilarities (24–30%, depending on contrast), together with Coffea canephora, Artocarpus spp., Albizia spp. and Eucalyptus spp., which collectively structure the compositional gradients across the three villages.

Composition of groups of use and key tree species across villages

Use-based group composition shifted markedly among villages (Fig. 5). In Ampahibe, forest species still accounted for almost half of all stems (48%), whereas this proportion dropped to 32% in Antsirakoraka and only 17% in Andampy. Cash-crops followed the opposite trend, increasing from 26% of all stems in Ampahibe to 56% in Antsirakoraka and 66% in Andampy, while fruit species contributed more moderately (26%, 13% and 17% of stems, respectively).

Within the cash-crop pool, Ampahibe exhibited multispecies portfolios combining S. aromaticum, V. planifolia, C. canephora and several minor cash crops. Clove, vanilla and coffee represented about 11%, 7.5% and 5.5% of all stems, respectively, and clove trees made up only 41% of the cash-oriented group. In Antsirakoraka, configurations were more clearly cash-oriented but still mixed, with clove contributing 36.5% of all stems and 66% of the cash-crop pool, together with substantial shares of vanilla (10.1%) and coffee (7.4%). In contrast, Andampy was strongly clove-dominated: clove trees alone accounted for 59% of all stems and about 90% of the cash-oriented group, while vanilla (4.6%) and coffee (1.6%) remained marginal. Species-level tests on density were fully consistent with this pattern: median clove density increased from 75 stems. ha⁻¹ in Ampahibe to 150 stems. ha⁻¹ in Antsirakoraka and 438 stems. ha⁻¹ in Andampy, with Ampahibe showing significantly lower densities than the two other villages (Kruskal–Wallis followed by pairwise Wilcoxon tests). By contrast, V. planifolia and C. canephora had very low median densities (0 stems. ha⁻¹ in all villages), and no significant differences among sites were detected for these species.

These patterns confirm contrasting agroforestry structures across villages, from diversified, fruit- and forest-rich systems in Ampahibe to simplified, clove-oriented systems in Andampy.

Cropping system distribution across villages

Consistent with the marked differences in species diversity and use-based composition observed among villages, the distribution of cropping systems differed markedly across villages (p = 0.003; Fig. 6). Ampahibe was overwhelmingly dominated by complex agroforests, which represented 89% of all plots, with the remaining 11% corresponding to diversified cash-oriented systems. Antsirakoraka exhibited a genuinely mixed pattern: complex agroforests accounted for 50% of plots, while diversified cash-oriented, clove-dominated and simple parkland systems each contributed 17%. In Andampy, complex agroforests became comparatively uncommon (28%), and the landscape was instead characterised by a high prevalence of clove-dominated systems (39%) together with a notable share of simple parklands and diversified cash-oriented systems (17% each). This gradient in system distribution mirrors the gradient in rice self-sufficiency capacity, with rice-abundant Ampahibe favouring diversified agroforests and rice-scarce Andampy dominated by cash-crop specialisation.

Village-scale differences in rice land availability drive agroforestry diversity patterns

Village-scale differentiation and rice autonomy

Our results showed that contrasts in clove-based system species composition and tree stand structure occurred primarily between villages, rather than along within-village toposequences. This village-scale differentiation was strongly associated with differences in access to irrigated lowlands and, consequently, with variation in potential rice self-sufficiency, which emerged as a major correlate of diversification–specialisation patterns.

The three villages represented contrasting accessible rice land and thus potential self-sufficiency levels (Table 1). Ampahibe showed an approximate rice surplus of + 15%, whereas Antsirakoraka and Andampy showed deficits of − 30% and − 90%, respectively. While these estimates are indicative rather than precise measures of household food security—they rely on regional agronomic parameters and do not capture local variation in yields, cropping intensity, market access, or off-farm income—they robustly reflect structural differences in rice availability across villages.

These differences in rice production capacity have direct economic implications. Villages facing rice deficits experience strong pressure to generate cash income to purchase staple food, constraining their capacity to maintain diversified agroforestry systems. Conversely, villages achieving rice self-sufficiency can afford greater flexibility in land-use decisions, maintaining species that provide longer-term benefits without jeopardising immediate food security (Droy et al. 2017).

These differences in rice land availability translated into clear diversification–specialisation trajectories of clove-based systems. In rice-surplus Ampahibe, clove-based systems were multispecies and structurally complex, with fruit and forest species accounting for 26% and 48% of stem density, respectively (Fig. 5), and 89% of plots classified as complex agroforests (Fig. 6). Antsirakoraka displayed intermediate configurations, with 50% of plots as complex agroforests and mixed representation of all four system types (Fig. 6). In contrast, rice-deficit Andampy exhibited strong specialisation, with clove accounting for 59% of stems (Fig. 5) and 39% of plots classified as clove-dominated systems (Fig. 6), alongside lower species richness (mean 7.0 ± 3.8 species per plot; Table 4) and minimal representation of fruit and forest tree species (17% each; Fig. 5).

Although our cross-sectional design does not allow causal inference, the convergence of independent indicators (species diversity, use-group composition, system-type distribution) consistently pointed to rice availability as a key factor shaping agroforestry diversity at the village scale. This finding aligns with agrarian research in Madagascar showing that irrigated rice underpins food security, labour allocation and social stability (Dabat et al. 2008; Droy et al. 2017).

From village patterns to household mechanisms

Although our cross-sectional design documents village-level patterns, the causal mechanism linking household rice autonomy to plot diversification was not directly tested. However, farm-level evidence from the region supports its plausibility. At the household scale, irrigated rice extent determines livelihood orientation: households achieving rice self-sufficiency mobilise agroforestry plots as complementary income sources, whereas those lacking lowlands depend on tree-crop revenues to purchase rice (Danthu et al., 2014; Andriatsitohaina et al., 2020; Mariel et al., 2021a).

The three villages represented distinct socio-ecological regimes reflecting contrasting rice production capacities (Table 1). In Ampahibe, extensive irrigated lowlands (1.09 ha per farm and 46% of the farmland area on average; Tonneau et al., 2018) provided high rice autonomy, enabling households to maintain diversified agroforests dominated by fruit and forest species. Clove contributes economically but does not substitute for staple-food production. Irrigated rice anchors food security, provides a stable base for livestock (Droy et al., 2017), and circulates as local currency—for repayment, wages, or exchange when cash is scarce (Houdaer, 2025). Farm-level income portfolios remain relatively diversified, with irrigated rice and livestock—grazed on lowlands during fallow or non-cultivated periods between rice cycles—contributing substantial shares alongside complementary tree crops (Ramalanjaona, 2025).

Antsirakoraka (0.47 ha per farm and 19% of the farmland area on average; Tonneau et al., 2018) showed intermediate rice self-sufficiency (–30% deficit) and mixed configurations. Households combined lowland rice production with diversified agroforests and cash-crop-oriented systems. Tree crops—clove, lychee, banana and vanilla—provide essential revenues and buffer lean periods, while lowland rice continues to secure a significant fraction of staple needs. During seasonal food shortages, households also rely on breadfruit and other food-producing trees (Mariel et al., 2021a, 2023). Tonneau et al. (2018) describe this reciprocity: tree crops fund rice purchases when needed, and rice sustains food security for cash-crop producers.

In Andampy (0.37 ha per farm and 13% of the farmland area on average; Tonneau et al., 2018), the near absence of irrigated rice (–90% deficit) structurally links household food security to clove revenues. Rainfed fields, complemented by maize and cassava, contribute to food security but cannot offset the absence of irrigation systems (Tonneau et al., 2018; Mariel et al., 2021b). Households rely structurally on clove production and essential oil to purchase rice and basic commodities (Danthu et al., 2014), with clove buds providing from 20 to 80% of total farm income and essential oil providing from 15 to 75% (Ramalanjaona, 2025). Agroforestry systems become central to livelihoods through ecological simplification: clove-dominated systems predominate (39% of plots; Fig. 6), clove accounts for 59% of stems (Fig. 5), and species richness is lowest (Table 4). Essential-oil distillation, which requires substantial fuelwood, may further reinforce removal of fruit and forest species (Danthu et al., 2020) and lower abundance of these species (Fig. 5).

In Ampahibe, despite 42% of landscape being covered by irrigated lowlands (Fig. 1; Table 1), household self-sufficiency remains fragile: many families sell harvests to meet immediate cash needs and later repurchase rice at higher prices—"structural self-sufficiency" (Mariel et al., 2021b). Here, wealth and social status appear to be more closely linked to rice self-sufficiency than to tree-crop holdings. We hypothesise that labour is therefore prioritised towards rice production, allowing diversified agroforests to be sustained but not actively invested in, which may explain the relatively low tree densities observed in Ampahibe (535 stems ha⁻¹; Table 4). Conversely, in Andampy, irrigated plots cover barely two months of consumption, forcing rice purchases with tree crop earnings. The clove tree has become both an economic foundation and a social marker—"wealth is measured by the number of productive clove trees" (Houdaer, 2025). Although clove harvesting yields up to fifteen times more income than irrigated rice (Fourcin et al., 2015), specialised systems are exposed to recurrent risks: periodic cyclones, occasional pest outbreaks, irregular bearing cycles, and market price fluctuations (Danthu et al., 2014). These risks do not occur annually but represent structural vulnerabilities that accumulate over time. Farmers adopt risk-spreading strategies—combining clove with fruit, timber, and food crops (Mariel et al., 2016)—but diversification scope is constrained where rice must be purchased.

Broader implications

This pattern aligns with broader evidence that staple-food security shapes diversification or specialisation trajectories in smallholder systems (Nyamayevu et al., 2024; Bosshard et al., 2025; Millet et al., 2025). Tittonell (2014) conceptualised smallholder heterogeneity as coexisting livelihood regimes shaped by resource endowments and market access. In the Iazafo landscape, access to irrigated rice acts as such a higher-level constraint, determining whether farmers maintain multifunctional agroforests (Ampahibe), combine staple and cash needs (Antsirakoraka), or specialise around cloves (Andampy).

These findings resonate with global tree cash-crop patterns. In Indonesia, oil palm expansion reinforced dependence on purchased rice (Budidarsono et al., 2013), while in Mexican shade-coffee systems, food self-sufficiency enables diversified agroforests (Perfecto et al., 2005). Where food autonomy is assured, farmers retain flexibility for multifunctional landscapes; where it is not, specialisation around high-value tree crops—oil palm, coffee, or clove—becomes essential to secure staple purchases.

Findings from Madagascar reinforce this dynamic. Andriatsitohaina et al. (2024) emphasise that agroforestry systems provide essential income and multifunctionality, yet their contribution to wellbeing remains constrained by structural vulnerabilities such as land scarcity, price volatility and livelihood fragility. Additionally, modelling work in northeastern Madagascar shows that when households depend structurally on cash-crop revenues to purchase rice, they become highly vulnerable to price fluctuations and harvest failures, creating potential "lock-in" dynamics in specialised systems (Celio et al., 2023).

Our results suggest that when rice autonomy is high, agroforestry diversity and multifunctionality can be sustained; when rice is structurally scarce, households rely on clove revenues to secure staple foods—a strategy that supports subsistence but drives ecological simplification and heightens exposure to climatic and market shocks.

Path dependency and rice-driven filtering in agroforestry diversification

The present mosaic of clove-based systems reflects the cumulative imprint of successive transformations: precolonial tavy-savoka cycles that generated widespread secondary vegetation, colonial coffee plantations established through coercive labour regimes (1897–1960), and post-coffee-crisis conversion to clove as farmers adapted to market collapse (Fremigacci, 1986; Blanc-Pamard and Ruf, 1992; Mariel et al., 2023). This historical succession—from shifting cultivation fallows to linear coffee systems to farmer-led clove expansion—produced the structural template of today's agroforestry mosaics and established clove as the regional economic pillar.

Contemporary diversification filtered by rice availability

Post-independence transformations added further layers of complexity. Studies by Arimalala et al. (2019) and Michel et al. (2021) document widespread enrichment of clove fields with fruit, medicinal and timber species as a strategy to buffer price volatility. Yet our results show that such diversification is no longer ubiquitous: its current expression is filtered primarily through village-scale rice production capacity. Farmers in Ampahibe, historically embedded in wide irrigated plains, maintain highly diversified agroforests with substantial representation of fruit and forest species (26% and 48% of stems, respectively; Fig. 5), with 89% of plots classified as complex agroforests (Fig. 6). In Antsirakoraka, systems exhibit intermediate enrichment reflecting mixed livelihood strategies, with 50% of plots as complex agroforests and the remaining plots distributed among the three other system types (Fig. 6). In contrast, Andampy—an upland region long associated with tavy and clove—exhibits the strongest specialisation, with clove accounting for 59% of stems (Fig. 5) and 39% of plots classified as clove-dominated systems (Fig. 6), supplying both clove buds and essential oil. This pattern echoes Andriatsitohaina et al. (2024), who report that when rice cannot meet household needs, farmers intensify cash-crop strategies to secure staple purchases.

This outcome reflects path dependency rooted in both land-use history and contemporary economic constraints. Osewold et al. (2022) showed in vanilla agroforests of northeastern Madagascar that past land use shapes present-day tree diversity, with forest-derived systems hosting richer species assemblages than fallow-derived ones. Similarly, in the Iazafo landscape, historical trajectories—from tavy fields to coffee plantations to clove agroforests—shape current species pools, management norms and regeneration pathways. However, our results reveal an additional filtering mechanism: ecological potential established by historical legacies is now channelled through contemporary rice availability, which determines whether households can afford to maintain diversified systems or must specialise to generate cash income.

Economic pressures override biophysical factors in species placement

Crucially, the spatial organisation of tree species within villages showed minimal response to topographic gradients. Despite documented species-environment associations—vanilla preferring moist lower slopes, coffee thriving under Albizia shade (Blanc-Pamard and Ruf, 1992; Mariel et al., 2022a)—our analyses detected no significant effect of hill position or orientation on species composition or on the distribution of major cash crops. The single marginal trend—higher clove density on west-facing slopes in Antsirakoraka (BH-adjusted p = 0.051)—did not reach statistical significance and was absent in both Ampahibe and Andampy (p = 0.83).

This pattern suggests that economic pressures increasingly override ecological niche-matching in species placement decisions. In rice-deficit Andampy (–90%), where household food security depends structurally on clove revenues, clove was planted across all topographic positions to maximise cash income, regardless of biophysical suitability. The marginal trend in Antsirakoraka (–30% deficit) may reflect intermediate conditions where households retain limited flexibility to adjust species placement to environmental gradients, though this capacity remains constrained. This interpretation aligns with Mariel's (2022) observation that tree species persist or are introduced primarily for their functional household value rather than ecological niche-matching: former coffee zones convert to clove regardless of shade conditions, savoka is enriched with diverse trees to buffer food shortages, and ecologically "suboptimal" occurrences persist because management decisions respond primarily to labour availability, food security needs, and market opportunities rather than to biophysical suitability.

Comparative perspectives on tree cash-crop path dependency

Comparable dynamics have unfolded in other tree cash-crop regions. In Zanzibar, a century of clove booms and busts created enduring dependence on a single export commodity shaped by state control, labour constraints and price volatility (Martin, 1991). In West and Central Africa, cocoa agroforests transitioned from utilising natural forest to diversified shade systems and, more recently, toward intensified monocultures under demographic and market pressures (Ruf and Siswoputranto, 1995; Ruf and Schroth, 2004; Jagoret et al., 2011; Michel et al., 2024). These cases reveal tree-crop landscapes as path-dependent socio-ecological systems, where past policies, market regimes and biophysical contexts continue to structure present possibilities for diversification or specialisation.

Within this broader lineage, the Iazafo region offers a distinctive articulation: historical introductions of cash crops (coffee, clove) provided the structural foundations of contemporary agroforestry systems, but their present heterogeneity is primarily filtered through rice self-sufficiency. Where irrigated lowlands secure food autonomy and labour flexibility (Ampahibe), farmers maintain multifunctional, diversified agroforests. Where rice is structurally scarce (Andampy), clove buds and essential oil become the principal means of accessing staple foods, driving simplification toward specialised systems. The resulting agroforestry gradient reflects the combined imprint of colonial legacies, postcolonial diversification practices, land-use history effects (sensu Osewold et al., 2022) and contemporary food-security constraints.

Research frontiers: from village patterns to within-farm heterogeneity

Our analysis clarifies how village-scale rice production capacity shapes broad contrasts in agroforestry composition. Yet landscape patterns reveal only part of the picture. Future research should articulate how these village-level gradients translate into finer-scale heterogeneity at the farm and plot levels.

First, within villages of intermediate rice deficit (such as Antsirakoraka), households with contrasting rice self-sufficiency levels likely coexist and adopt fundamentally different agroforestry strategies. Droy et al. (2017) demonstrate that livelihood strategies hinge on the interplay between land endowments and irrigated lowland availability. Rice-deficit households may concentrate their upland plots on intensive clove production to maximise cash income for staple purchases, while households achieving rice self-sufficiency may maintain less-intensified land reserves—fallows, parklands, or enriched secondary vegetation—that provide timber, fruit, or future conversion opportunities without jeopardising immediate food security. Integrating our village-scale typology with farm-level surveys quantifying household rice balances and upland plot composition is therefore essential to explain why complex agroforests and clove-dominated systems coexist within the same village.

Second, even within farms, individual upland plots often exhibit pronounced compositional heterogeneity. At the farm scale, households typically manage multiple plots with different configurations: some plots may be clove-dominated, others may retain parkland structures or be maintained as enriched fallows. This among-plot variability within farms likely reflects differences in plot acquisition history, soil fertility, topographic position, and asynchronous enrichment trajectories. Mariel et al. (2022a) showed that farmers differentiate management strategies according to topographic position within the farm territory, concentrating fruit and shade trees on fertile lower slopes and clove on upper, less fertile zones. Understanding how households allocate management effort and species composition across their portfolio of plots—rather than treating "the farm" as a homogeneous unit—is essential for interpreting observed landscape patterns.

This multi-scalar perspective recognises that agroforestry diversity cannot be understood from village patterns alone. Village-scale rice availability sets broad constraints on the range of feasible strategies, but these constraints are filtered through household-level resource endowments (land, labour, capital) and materialised in plot-level management decisions shaped by soil conditions, acquisition history, and labour availability. Integrating these nested scales—village, household, and plot—is essential to fully understand the mechanisms generating observed patterns of diversification and specialisation in Madagascar's clove-dominated landscapes.

This study reveals that village-level access to irrigated lowlands is strongly associated with agroforestry diversification–specialisation trajectories in eastern Madagascar's clove-dominated landscapes. By combining a structural typology of agroforestry systems with a quantitative proxy of rice production capacity, we showed that village-scale differences in rice land availability align consistently with compositional and structural patterns, whereas within-village topography had no detectable effect on species distributions.

Across the Iazafo gradient, agroforestry diversity declined systematically with rice deficit. In villages with abundant irrigated lowlands (Ampahibe), complex agroforests integrating fruit, forest and cash-crop species persisted. In upland villages with severe rice deficits (Andampy), households relied structurally on clove revenues to purchase staple foods, driving ecological simplification toward clove-dominated systems. Although our cross-sectional design precludes causal inference, the convergence of independent indicators—species diversity, functional- group composition, system-type distribution, and multivariate ordination— consistently pointed to rice land availability as a key structuring factor. These results suggest that food security and perennial-crop economies co-evolve: clove complements diversification where rice is sufficient, but substitutes for food production where rice is structurally scarce.

Historical legacies—colonial coffee plantations, farmer-led clove expansion, post-independence enrichment—continue to shape present-day mosaics, yet their expression appears filtered primarily through rice self-sufficiency. Contemporary configurations reflect long-term interactions between land-use history, household resource endowments and the functional value of species, rather than ecological sorting alone.

While these findings are derived from three villages and require validation across broader geographic and socio-ecological gradients, they suggest that strategies jointly supporting irrigated lowlands, diversified enrichment and improved clove management may offer robust pathways for sustaining ecological complexity, livelihood security and resilience in Madagascar's clove-producing regions. Future research integrating household-level data on rice balance and farm-level plot heterogeneity is essential to confirm and refine these patterns.

Funding

This work was supported by the French National Research Agency (ANR) under the BiodivClo project ("Biodiversity for the resilience of clove-based agroecosystems in Madagascar"), referred to as ANR-22-CE32-0002.

Competing Interests

The authors have no competing financial or non-financial interests to declare that are relevant to the content of this article.

Author Contributions

Manon Desmurs: Conceptualisation, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing – review & editing.

Julien Sarron: Funding acquisition, Investigation, Methodology, Project administration, Supervision, Validation, Writing – review & editing.

Elisa Rahantarivelo: Investigation, Validation, Writing – review & editing.

Pascal Danthu: Conceptualization, Supervision, Writing – review & editing.

Éric Malézieux: Conceptualization, Supervision, Writing – review & editing.

Isabelle Michel: Formal analysis, Methodology, Supervision, Writing – review & editing.

All authors read and approved the final manuscript.

Ethics Approval

This research did not involve human participants, clinical trials, or experimental work with animals. Ethical approval was therefore not required.

Consent to Participate

Not applicable.

Consent for Publication

All authors consent to the publication of this manuscript.

Data Availability

The datasets generated and analysed during this study (tree-inventory data and spatial layers) are available from the corresponding author upon reasonable request. Geospatial layers and R scripts used for cropping system typology, diversity analysis and statistical tests have been archived in an open repository (link to be provided at submission). A DOI will be supplied in the final version.

Acknowledgements

We express our gratitude to the farmers and local authorities of Ampahibe, Antsirakoraka and Andampy for their hospitality and help in data collection. Fieldwork was conducted within the BiodivClo project ("Biodiversity for the resilience of clove-based agroecosystems in Madagascar"), coordinated by CIRAD (UPR HortSys) in collaboration with FOFIFA (Madagascar). We warmly thank Zozo Raveloson and Jenny Tianaina (Institut Supérieur de Technologie Régional de la Côte Est – Université de Toamasina) for their invaluable support during fieldwork. We are grateful to David Marsh for his careful proofreading of the English manuscript.

Anderson, M.J., 2001. A new method for non‐parametric multivariate analysis of variance. Austral Ecol. 26, 32–46. https://doi.org/10.1111/j.1442-9993.2001.01070.pp.x
Andriatsitohaina, R.N.N., Celio, E., Llopis, J.C., Rabemananjara, Z.H., Ramamonjisoa, B.S., Grêt-Regamey, A., 2020. Participatory Bayesian network modeling to understand driving factors of land-use change decisions: insights from two case studies in northeast Madagascar. J. Land Use Sci. 15, 69–90. https://doi.org/10.1080/1747423X.2020.1742810
Andriatsitohaina, R.N.N., Laby, P., Llopis, J.C., Martin, D.A., 2024. Agroforestry in Madagascar: past, present, and future. Agrofor. Syst. 98, 1659–1680. https://doi.org/10.1007/s10457-024-00975-y
Arimalala, N., Penot, E., Michels, T., Rakotoarimanana, V., Michel, I., Ravaomanalina, H., Roger, E., Jahiel, M., Leong Pock Tsy, J.-M., Danthu, P., 2019. Clove based cropping systems on the east coast of Madagascar: how history leaves its mark on the landscape. Agrofor. Syst. 93, 1577–1592. https://doi.org/10.1007/s10457-018-0268-9
Bates, D., Mächler, M., Bolker, B., Walker, S., 2015. Fitting Linear Mixed-Effects Models Using lme4. J. Stat. Softw. 67, 1–48. https://doi.org/10.18637/jss.v067.i01
Be, A.F., Sarron, J., 2023. Reconnaissance de la zone d’étude Antsirakoraka pour le projet ANR-BiodivClo. Rapport de mission Antsirakoraka (district de Vavatenina) (Rapport de mission). CIRAD, Montpellier, France.
Beillouin, D., Ben‐Ari, T., Malézieux, E., Seufert, V., Makowski, D., 2021. Positive but variable effects of crop diversification on biodiversity and ecosystem services. Glob. Change Biol. 27, 4697–4710. https://doi.org/10.1111/gcb.15747
Berger, W.H., Parker, F.L., 1970. Diversity of planktonic foraminifera in deep-sea sediments. Science 168, 1345–1347.
Blanc-Pamard, C., Ruf, F., 1992. La transition caféière: Côte Est de Madagascar, Collection documents systèmes agraires. Département Systèmes Agroalimentaires et Ruraux du CIRAD, Montpellier.
Bosshard, E., Carter, H., Aluso, L., Chumba, R., Kaiser‐Bunbury, C.N., Kettle, C.J., Nuno, A., 2025. Understanding smallholder decision‐making to increase farm tree diversity: Enablers and barriers for forest landscape restoration in Western Kenya. People Nat. pan3.10774. https://doi.org/10.1002/pan3.10774
Bray, J.R., Curtis, J.T., 1957. An ordination of the upland forest communities of southern Wisconsin. Ecol. Monogr. 27, 326–349.
Budidarsono, S., Susanti, A., Zoomers, A., 2013. Oil Palm Plantations in Indonesia: The Implications for Migration, Settlement/Resettlement and Local Economic Development, in: Fang, Z. (Ed.), Biofuels - Economy, Environment and Sustainability. InTech. https://doi.org/10.5772/53586
Celio, E., Andriatsitohaina, R.N.N., Llopis, J.C., Gret-Regamey, A., 2023. Assessing farmers’ income vulnerability to vanilla and clove export economies in northeastern Madagascar using land-use change modelling. J. Land Use Sci. 18, 55–83. https://doi.org/10.1080/1747423X.2023.2168778
Clarke, K.R., 1993. Non‐parametric multivariate analyses of changes in community structure. Aust. J. Ecol. 18, 117–143. https://doi.org/10.1111/j.1442-9993.1993.tb00438.x
Dabat, M.-H., Jenn-Treyer, O., Razafimandimby, S., Bockel, L., 2008. L’histoire inachevée de la régulation du marché du riz à Madagascar. Économie Rurale 75–89. https://doi.org/10.4000/economierurale.535
Dandoy, G., 1973. Territoires et Économies Villageoises de La Région de Vavatenina (Côte Orientale Malgache), Atlas Des Structures Agraires à Madagascar. MOUTON & CO, Paris.
Danthu, P., Michel, I., Carrière, S.M., Labeyrie, V., Rakouth, B., Sarron, J., Mariel, J., Lasserre, D., Penot, E., 2022. Coming from elsewhere: the preponderance of introduced plant species in agroforestry systems on the east coast of Madagascar. Agrofor. Syst. 96, 697–716. https://doi.org/10.1007/s10457-022-00732-z
Danthu, P., Penot, E., Ranoarisoa, K., Rakotondravelo, J., Michel, I., Tiollier, M., Michels, T., Normand, F., Razafimamonjison, G., Fawbush, F., Jahiel, M., 2014. The clove tree of Madagascar: A success story with an unpredictable future. Bois Forets Trop. 68, 83–96.
Danthu, P., Sarron, J., Penot, E., Mariel, J., Razafimaharo, V., Michel, I., 2024. The agroforests of the east coast of Madagascar. Trop. For. Issues. https://doi.org/10.55515/HDZZ3352
Danthu, P., Simanjuntak, R., Fawbush, F., Leong Pock Tsy, J.-M., Razafimamonjison, G., Abdillahi, M.M., Jahiel, M., Penot, E., 2020. The clove tree and its products (clove bud, clove oil, eugenol): Prosperous today but what of tomorrow’s restrictions? Fruits 75, 224–242. https://doi.org/10.17660/th2020/75.5.5
Droy, I., Rabevohitra, B.N., Bidou, J.-É., 2017. L’encastrement des vulnérabilités : exemple du « pays du girofle » sur la côte orientale de Madagascar: Mondes En Dév. n° 180, 69–86. https://doi.org/10.3917/med.180.0069
Fourcin, C., Penot, E., Michel, I., Danthu, P., Jahiel, M., 2015. Contribution du giroflier à la sécurité alimentaire des ménages agricoles dans la région de Fénérive-Est, Madagascar Modélisation économique et analyse prospective. CIRAD 56. https://doi.org/10.13140/RG.2.1.3020.6880
Fremigacci, J., 1986. Les difficultés d’une politique coloniale : le café de Madagascar à la conquête du marché français (1930-1938). Rev. Fr. Hist. O.-m. 73, 143–166. https://doi.org/10.3406/outre.1986.2522
Houdaer, A., 2025. La forêt des invisibles. Usages, échanges et attachements aux plantes et ressources des agroforêts de la côte Est malgache (Mémoire). Institut Agro Montpellier, Montpellier, France.
International Monetary Fund. African Dept., 2025. Republic of Madagascar: Selected Issues. IMF Staff Ctry. Rep. 2025, 1. https://doi.org/10.5089/9798229004114.002
Jagoret, P., Michel-Dounias, I., Malézieux, E., 2011. Long-term dynamics of cocoa agroforests: a case study in central Cameroon. Agrofor. Syst. 81, 267–278. https://doi.org/10.1007/s10457-010-9368-x
Jose, S., 2009. Agroforestry for ecosystem services and environmental benefits: an overview. Agrofor. Syst. 76, 1–10. https://doi.org/10.1007/s10457-009-9229-7
Kruskal, J.B., 1964. Multidimensional Scaling by Optimizing Goodness of Fit to a Nonmetric Hypothesis. Psychometrika 29, 1–27. https://doi.org/10.1007/BF02289565
Kuznetsova, A., Brockhoff, P.B., Christensen, R.H.B., 2017. lmerTest Package: Tests in Linear Mixed Effects Models. J. Stat. Softw. 82, 1–26. https://doi.org/10.18637/jss.v082.i13
Mariel, J., 2022. Dynamiques de l’agrobiodiversité dans les paysages agroforestiers de Vavatenina à Madagascar : études des pratiques, des connaissances paysannes et des réseaux d’échange dans un contexte de changements (Dissertation). Univ. de Montpellier, Montpellier.
Mariel, J., Carrière, S.M., Penot, E., Danthu, P., Rafidison, V., Labeyrie, V., 2021a. Exploring farmers’ agrobiodiversity management practices and knowledge in clove agroforests of Madagascar. People Nat. 3, 914–928. https://doi.org/10.1002/pan3.10238
Mariel, J., Freycon, V., Randriamalala, J., Rafidison, V., Labeyrie, V., 2022a. Local Knowledge of the Interactions between Agrobiodiversity and Soil: A Fertile Substrate for Adapting to Changes in the Soil in Madagascar? J. Ethnobiol. 42, 180–197. https://doi.org/10.2993/0278-0771-42.2.180
Mariel, J., Penot, É., Danthu, P., 2021b. Les systèmes agroforestiers dans les stratégies paysannes de résilience à Madagascar. Économie Rurale 43–63. https://doi.org/10.4000/economierurale.9104
Mariel, J., Penot, E., Labeyrie, V., Herimandimby, H., Danthu, P., 2023. From shifting rice cultivation (tavy) to agroforestry systems: a century of changing land use on the East Coast of Madagascar. Agrofor. Syst. 97, 415–431. https://doi.org/10.1007/s10457-022-00761-8
Mariel, J., Penot, E., Michel, I., Danthu, P., 2016. Analyse des systèmes agroforestiers du territoire de Vavatenina à Madagascar : stratégies de mise en valeur, perceptions paysannes et résilience, in: 10èmes Journées de recherches en sciences sociales (JRSS). Paris, France, p. 22.
Mariel, J., Sanchez, I., Verzelen, N., Massol, F., Carrière, S.M., Labeyrie, V., 2022b. Translocality in the Circulation of Plants and Knowledge: Which Perspectives for Agricultural Systems Resilience? SSRN Electron. J. https://doi.org/10.2139/ssrn.4130767
Martin, P.J., 1991. The Zanzibar clove industry. Econ. Bot. 45, 450. https://doi.org/10.1007/BF02930706
Michel, I., Blanco, J., Manga Essouma, F., Carrière, S.M., 2024. Complex cocoa agroforestry systems shaped within specific socioeconomic and historical contexts in Africa: Lessons from Cameroonian farmers. Agric. Syst. 221, 104111. https://doi.org/10.1016/j.agsy.2024.104111
Michel, I., Lobietti, M., Danthu, P., Penot, E., Velonjara, F., Jahiel, M., Michels, T., 2021. Agroforestry innovation by smallholders facing uncertainty: The case of clove-based cropping systems in Madagascar. Eur. J. Agron. 123, 126218. https://doi.org/10.1016/j.eja.2020.126218
Millet, C.P., Jeune, W., Guervil, J.S., St Armand, L.A., Amazan, J.F., Duval, G., Jean Louis, R.B., Robert, B., Poncet, V., Allinne, C., 2025. Ecosystem service bundles associated with agrobiodiversity in agroforestry systems: A case study of two coffee-growing regions of Haiti. Ecosyst. Serv. 76, 101782. https://doi.org/10.1016/j.ecoser.2025.101782
Nyamayevu, D., Nyagumbo, I., Chiduwa, M., Liang, W., Li, R., 2024. Understanding Crop Diversification Among Smallholder Farmers: Socioeconomic Insights from Central Malawi. Sustainability 16, 9078. https://doi.org/10.3390/su16209078
Ogle, D.H., Doll, J.C., Wheeler, A.P., Dinno, A., 2023. FSA: Simple Fisheries Stock Assessment Methods.
Oksanen, J., Simpson, G.L., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O’Hara, R.B., Solymos, P., Stevens, M.H.H., Szoecs, E., Wagner, H., Barbour, M., Bedward, M., Bolker, B., Borcard, D., Carvalho, G., Chirico, M., Caceres, M.D., Durand, S., Evangelista, H.B.A., FitzJohn, R., Friendly, M., Furneaux, B., Hannigan, G., Hill, M.O., Lahti, L., McGlinn, D., Ouellette, M.-H., Cunha, E.R., Smith, T., Stier, A., Braak, C.J.F.T., Weedon, J., 2022. vegan: Community Ecology Package.
Osewold, J., Korol, Y., Osen, K., Soazafy, M.R., Rabemanantsoa, T., Martin, D.A., Wurz, A., Hölscher, D., 2022. Support trees in vanilla agroforests of Madagascar: diversity, composition and origin. Agrofor. Syst. 96, 717–730. https://doi.org/10.1007/s10457-022-00733-y
Perfecto, I., Vandermeer, J., Mas, A., Pinto, L.S., 2005. Biodiversity, yield, and shade coffee certification. Ecol. Econ. 54, 435–446. https://doi.org/10.1016/j.ecolecon.2004.10.009
Pielou, E.C., 1966. The measurement of diversity in different types of biological collections. J. Theor. Biol. 13, 131–144. https://doi.org/10.1016/0022-5193(66)90013-0
Ramalanjaona, A., 2025. Caractérisation technico-économique des exploitations agricoles et des systèmes à base de giroflier à Madagascar (Mémoire). ISTOM, Angers, France.
Ruf, F., Schroth, G., 2004. Chocolate forests and monocultures - an historical review of cocoa growing and its conflicting role in tropical deforestation and forest conservation, in: Schroth, G., Da Fonseca, G.A.B., Harvey, C.A., Gascon, C., Vasconcelos, H.L., Izac, A.-M.N. (Eds.), Agroforestry and Biodiversity Conservation in Tropical Landscapes. Washington D.C., pp. 107–134.
Ruf, F., Siswoputranto, P.S. (Eds.), 1995. Cocoa cycles: the economics of cocoa supply. Woodhead, Abington.
Schroth, G., Harvey, C.A., Vincent, G., 2004. Complex agroforests : their structure, diversity and potential role in landscape conservation, in: Schroth, Götz, Da Fonseca, G.A.B., Harvey, Celia A., Gascon, C., Vasconcelos, H.L., Izac, A.-M.N. (Eds.), Agroforestry and Biodiversity Conservation in Tropical Landscapes. Washington D.C., pp. 227–260.
Shannon, C.E., 1948. A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423.
Simpson, E.H., 1949. Measurement of Diversity. Nature 163, 688–688. https://doi.org/10.1038/163688a0
Stratégie Nationale de Développement Rizicole III: 2022–2030 (Rapport national de politique publique), 2022. Ministère de l’Agriculture et de l’Élevage (MINAE), République de Madagascar, Antananarivo, Madagascar.
Terasaki Hart, D.E., Yeo, S., Almaraz, M., Beillouin, D., Cardinael, R., Garcia, E., Kay, S., Lovell, S.T., Rosenstock, T.S., Sprenkle-Hyppolite, S., Stolle, F., Suber, M., Thapa, B., Wood, S., Cook-Patton, S.C., 2023. Priority science can accelerate agroforestry as a natural climate solution. Nat. Clim. Change 13, 1179–1190. https://doi.org/10.1038/s41558-023-01810-5
Tittonell, P., 2014. Livelihood strategies, resilience and transformability in African agroecosystems. Agric. Syst. 126, 3–14. https://doi.org/10.1016/j.agsy.2013.10.010
Tonneau, J.-P., Andrianarivelo, M., Bélières, J.-F., Bouvet, J.-M., Herimandimby, H., Queste, J., Rakotondrainibe, J.H., Razafindrianilana, T., 2018. Études préparatoires du Projet d’Agriculture Durable par une Approche Paysage (PADAP) Paysages d’Iazafo et de Soanierana Ivongo Région Analanjirofo Madagascar. (Rapport d’expertise). CIRAD, Montpellier, France.
Tscharntke, T., Clough, Y., Bhagwat, S.A., Buchori, D., Faust, H., Hertel, D., Hölscher, D., Juhrbandt, J., Kessler, M., Perfecto, I., Scherber, C., Schroth, G., Veldkamp, E., Wanger, T.C., 2011. Multifunctional shade-tree management in tropical agroforestry landscapes - a review: Multifunctional shade-tree management. J. Appl. Ecol. 48, 619–629. https://doi.org/10.1111/j.1365-2664.2010.01939.x

No competing interests reported.

Download PDF

Reviews received at journal
08 Jan, 2026
Reviews received at journal
06 Jan, 2026
Reviewers agreed at journal
26 Dec, 2025
Reviewers agreed at journal
23 Dec, 2025
Reviewers agreed at journal
22 Dec, 2025
Reviewers invited by journal
22 Dec, 2025
Editor assigned by journal
19 Dec, 2025
Submission checks completed at journal
19 Dec, 2025
First submitted to journal
18 Dec, 2025

You are reading this latest preprint version

How rice land availability shapes diversification–specialisation gradients in Madagascar's clove landscapes

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Study area

Sampling design and field surveys

Hillslope transects

Vegetation plots and woody-inventory protocol

Understorey and associated crops

Processing of floristic and structural data

Groups of uses

System-structure metrics

Alpha diversity and dominance

Typology of clove cropping system

Statistical analyses

Results

Four clove cropping systems along a diversification–specialisation gradient

Simple parkland

Clove-dominated system

Diversified cash-oriented system

Complex agroforests

Within-village variation of diversity along hill section

Alpha diversity along hill sections

Composition of groups of use and key tree species along hills

Distribution of cropping-system types along hill positions

Differences between villages

Floristic and structural diversity across villages

Composition of groups of use and key tree species across villages

Cropping system distribution across villages

Discussion

Village-scale differences in rice land availability drive agroforestry diversity patterns

Village-scale differentiation and rice autonomy

From village patterns to household mechanisms

Broader implications

Path dependency and rice-driven filtering in agroforestry diversification

Contemporary diversification filtered by rice availability

Economic pressures override biophysical factors in species placement

Comparative perspectives on tree cash-crop path dependency

Research frontiers: from village patterns to within-farm heterogeneity

Conclusion

Declarations

Funding

Competing Interests

Author Contributions

Ethics Approval

Consent to Participate

Consent for Publication

Data Availability

Acknowledgements

References

Additional Declarations

Status:

Version 1