Tree diversity and composition of cocoa agroforests in Cote d'Ivoire: a regional ecological mosaic, implications for cocoa landscapes sustainable management

Cocoa (Theobroma cacao L.) is grown in Cote d'Ivoire (the world's leading producer), almost always in association with trees, yet these cocoa-based agroforestry systems (cocoa AFS) remain characterized largely descriptively, without a robust inferential framework or modern diversity indices. We analyse a survey database covering 474 farms and 103 localities across the three main cocoa-producing zones of Cote d'Ivoire, locally known as " loops " (East/South-East, Centre/Centre-West, South/South-West; surveys 2013-2016), applying a renewed ecological and statistical framework: Hill numbers (q = 0, 1, 2), individual-based Hurlbert rarefaction, multiplicative alpha/beta/gamma diversity partitioning, ordination (PCoA) and PERMANOVA on Bray-Curtis matrices, Ward's hierarchical clustering for a data-driven floristic typology, multivariate MANOVA, and linear mixed models. Across 99 adequately sampled localities (>10 stems), mean taxonomic richness and the Shannon index differed significantly among zones (MANOVA: Wilks' = 0.350; F8,180 = 15.5; p < 0.001), including after standardizing sampling effort by rarefaction (F = 8.72; p < 0.001). Floristic composition differed significantly by zone (PERMANOVA, pseudo-F = 2.85; p = 0.001), though with uneven within-group dispersion (PERMDISP, p = 0.024) that calls for caution in interpretation. The Centre/Centre-West zone showed the highest beta diversity (inter-farm floristic turnover = 16.4, versus 7.2 and 2.2 in the other two zones), revealing a highly heterogeneous mosaic of individual agroforestry choices rather than a homogeneous regional system. A typological classification based on floristic composition distinguishes four cocoa-AFS profiles, ranging from dense, diverse stands to impoverished systems with low densities of associated trees, partially but not fully overlapping with the zonal boundaries. At the scale of the three loops (n = 409 farms with both age and yield recorded), cocoa yield varied significantly by orchard age class (Kruskal-Wallis, H = 18.2; p < 0.001), following a bell-shaped profile peaking around 20-35 years, consistent with the intermediate-age yield optimum suggested qualitatively by the source study; this relationship, robust to log-transformation of yield and to the exclusion of extreme values, nonetheless remains of small magnitude (R2=1.5%) and confers no out-of-sample predictive power under cross-validation, illustrating the need to distinguish statistical significance from predictive utility. This reanalysis provides a reproducible and transferable statistical framework for characterizing West African cocoa AFS, and highlights the potential value of high beta diversity as a possible indicator of landscape resilience.