Socioeconomic and biophysical drivers of cocoa yield variability in smallholder farms in Côte d’Ivoire Authors: Nanga Yéo, Demeango Serge Zon, Ebagnerin Jérôme Tondoh Published: 2026 ABSTRACT Côte d’Ivoire’s cocoa sector faces persistent low and variable yields, influenced by social, agronomic, and ecological factors, which in turn drive agricultural expansion and deforestation. Integrated studies on yield determinants remain limited. This study examines these drivers across five sub-prefectures in two climatic zones. Data from 158 cocoa producers on farm characteristics, management practices, and perceived biophysical constraints were analyzed using descriptive statistics and a generalized linear mixed model (GLMM) with a Gamma distribution and log-link, incorporating study zones as random effects. Yields varied from 372 to 812 kg ha⁻¹ year⁻¹ across sub-prefectures, with no statistically significant differences between climatic zones. Three factors significantly increased yields: mineral fertilization (+ 29%, P = 0.036), aligned cocoa tree planting (+ 51%, P = 0.01), and the use of identified planting material (+ 23%, P = 0.04). Farmer age (-52%, P = 0.002) was negatively associated with productivity. Key production constraints identified by farmers included insect damage (44.9%), black pod disease (43.0%), low soil fertility (31.0%), cocoa tree mortality (27.2%), and drought (20.9%), with variation across sites. Sustainable yield improvement requires integrated interventions combining technical innovations, improved access to inputs, strengthened extension services, and generational renewal. Tailoring measures to local conditions is essential to maximize productivity and reduce deforestation pressures linked to agricultural expansion.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-42681-y. Introduction Cocoa (Theobroma cacao L.) value chains play a critical role in the global economy, particularly in West Africa, which supplies more than 70% of global production1. Côte d’Ivoire has maintained its position as the world’s leading producer for several decades, accounting for roughly 43% of global supply. National production exceeded two million metric tons during the 2020/21 and 2022/23 seasons but fell below this level in the 2023/24 season2. The sector is a cornerstone of the Ivorian economy, contributing over 15% of agricultural GDP and directly supporting the livelihoods of more than one million smallholder farmers, mainly located in the southern and western forest zones3. Despite this apparent success, the expansion of cocoa production areas masks deep-rooted structural vulnerabilities. In both Côte d’Ivoire and Ghana, production gains have been achieved largely through area expansion—often into forested lands—rather than yield improvements4–8. Current yields remain low, typically 400–600 kg ha⁻¹ year⁻¹, far below the agronomic potential of over 2,000 kg ha⁻¹ year⁻¹ under optimal conditions9–11. To offset low yields, many farmers continue to encroach on protected forests, thereby accelerating deforestation, biodiversity loss, and soil degradation12–16. This expansion-based model is increasingly unsustainable due to land scarcity and the continued deterioration of smallholder livelihoods. Against this backdrop, enhancing cocoa productivity through sustainable intensification has become imperative. Achieving this goal requires a comprehensive understanding of the agronomic, climatic, and socioeconomic constraints that limit yields9,17,18. From a biophysical perspective, key challenges widely reported across cocoa-growing regions include aging plantations, declining soil fertility, pest and disease pressures, and increasing climatic variability19–21. Cocoa trees typically begin to senesce after about 25 years, showing reduced vigor, lower pod set, and smaller beans9,22. This biological decline is often exacerbated by progressive soil nutrient depletion, which increases plant vulnerability in the absence of adequate fertilization5,23. Furthermore, the lack of essential agronomic practices—such as regular pruning and pest management— intensifies yield losses. For example, in Ghana, black pod disease (Phytophthora spp.) can cause yield losses exceeding 30% in the absence of proper phytosanitary control24,25. Insect pests also cause substantial damage throughout the production cycle26,27, while climatic stressors—including prolonged droughts and erratic rainfall—further impair flowering and pod development and create favorable conditions for fungal outbreaks5,26,28,29. At the plot level, farmers adopt diverse management strategies shaped by their local knowledge and economic capacities. However, widespread poverty, limited access to credit and agricultural inputs, and insufficient technical support remain major barriers to productivity improvements8,30–32. Recent studies indicate that many cocoa farmers in Côte d’Ivoire and Ghana live below the international poverty line of USD 2.12/day (PPP)8,32. Most are unable to earn a living income, which severely constrains their capacity to invest in fertilizers, pesticides, or hired labor32,33. As a result, some farmers shift to alternative cash crops such as rubber or oil palm, while rural outmigration—particularly among youth—further reduces the agricultural labor force and jeopardizes the sector’s long-term sustainability34–37. Although Ghana and several other cocoa-producing countries have been the subject of extensive socioeconomic research38–43, Côte d’Ivoire—despite being the world’s largest producer—still lacks up-to-date and comprehensive socioeconomic data on cocoa farming. Existing information is often outdated or only indirectly related to productivity44–47. This gap is particularly concerning considering current transformations affecting the sector, including the agroecological transition, the roll-out of deforestation-free cocoa initiatives, and compliance with the European Union Deforestation Regulation (EUDR). Past interventions to improve productivity have frequently emphasized technical solutions, such as soil fertility management or agroforestry practices48–52, while neglecting the broader socioeconomic context and farmers’ own perceptions. Yet, farmer decisions are shaped not only by available resources but also by personal experiences and perceived constraints53–55. Recent evidence further shows that the adoption of agricultural innovations strongly depends on farmers’ socioeconomic profiles and farm characteristics30,56–59. In this context, generating updated, spatially disaggregated socioeconomic data across cocoa production zones in Côte d’Ivoire is urgently needed. Such information is critical for designing interventions that are both context-specific and broadly scalable, thereby supporting a sustainable and inclusive transformation of the cocoa sector. To address these challenges, this study investigates the biophysical and socioeconomic drivers of yield variability in cocoa farming systems across major production regions of Côte d’Ivoire, while also documenting farmers’ perceptions of key biophysical constraints to production. Specifically, the study addresses the following research questions: (i)What are the socio-demographic characteristics of cocoa farmers and the agricultural practices they implement across five major cocoa-growing sites?(ii)How does cocoa yield vary across study zones and ecological sectors?(iii)What are the key agronomic and socioeconomic determinants of cocoa yield?(iv)What are the main biophysical constraints to cocoa production as perceived by farmers? (i)What are the socio-demographic characteristics of cocoa farmers and the agricultural practices they implement across five major cocoa-growing sites? What are the socio-demographic characteristics of cocoa farmers and the agricultural practices they implement across five major cocoa-growing sites? (ii)How does cocoa yield vary across study zones and ecological sectors? How does cocoa yield vary across study zones and ecological sectors? (iii)What are the key agronomic and socioeconomic determinants of cocoa yield? What are the key agronomic and socioeconomic determinants of cocoa yield? (iv)What are the main biophysical constraints to cocoa production as perceived by farmers? What are the main biophysical constraints to cocoa production as perceived by farmers? Materials and methods Study area description This study was conducted in the cocoa-growing regions of southern Côte d’Ivoire. Five sub-prefectures were selected as survey sites: Maféré, Grand-Béréby, Divo, Yakassé-Attobrou (Biébi), and Guibéroua (Fig. 1). These sites are part of the Cocoa4Future project’s network of fifteen research clusters (https://www.cocoa4future.org), selected to represent a pedoclimatic gradient that reflects the agroecological diversity of cocoa production systems in the country. From an edaphic perspective and based on the FAO soil classification system as represented in the Harmonized World Soil Database (HWSD), three major soil types characterize the study areas: Xanthic Ferralsols, Orthic Acrisols, and Ferric Acrisols. According to HWSD-derived data, Ferric Acrisols predominate in the Guibéroua and Divo regions, whereas Xanthic Ferralsols are dominant in Maféré and Grand-Béréby. The Yakassé-Attobrou (Biébi) area is primarily composed of Orthic Acrisols. These soils differ markedly in their physico-chemical properties, which are critical determinants of fertility and crop performance. Xanthic Ferralsols are highly weathered tropical soils, typically yellow to red in color, rich in iron oxides and kaolinitic clays. They possess favorable physical properties such as good structure and drainage but are naturally nutrient-poor and exhibit high phosphorus fixation due to intense leaching in humid environments60. Orthic Acrisols are acidic soils with pronounced clay accumulation in the subsoil (B horizon) and low base saturation (< 50%). They typically occur under humid tropical climates and are associated with forest or savannah vegetation. Ferric Acrisols, a subtype of Acrisols, are enriched in iron oxides, often showing ferruginous concretions and a reddish hue. Although less leached than Ferralsols, they remain poor in essential nutrients61. Climatic conditions also vary across the study sites, which span two major agroecological zones. Maféré, Grand-Béréby, and Biébi fall within the ombrophilous (moist evergreen) rainforest zone of the southeastern and southwestern coastal regions, while Divo and Guibéroua lie in the humid semi-deciduous forest zone extending from the south-central to west-central areas62. The ombrophilous zone receives 1,500–2,000 mm of annual rainfall with relatively weak seasonal variation, whereas the semi-deciduous zone receives 1,000–1,500 mm, marked by a distinct alternation between dry and rainy seasons. Mean annual temperatures range between 25 and 26 °C in both zones, with slightly higher values recorded in the ombrophilous areas. Elevation varies from 100 to 200 m above sea level in coastal zones to 200–300 m in inland areas63. Ethical approval and informed consent All procedures implemented in this study were conducted in accordance with applicable guidelines and ethical standards. Before the start of data collection, authorization to carry out the surveys was obtained from the traditional authorities of each study village. Subsequently, the purpose of the study and the principles governing the ethical use of collected information were clearly explained to all participating cocoa farmers prior to the interviews. Participants were informed that their participation was entirely voluntary and that they could withdraw from the survey at any stage, decline to answer specific questions, or temporarily suspend the interview without any consequences. Only farmers who confirmed their understanding of these conditions and agreed to participate were interviewed, and verbal informed consent was obtained from all respondents. The survey did not involve biomedical research, clinical procedures, biological sampling, experimental interventions, or vulnerable populations. It also did not involve the collection of sensitive personal data, such as health information, political or religious beliefs, or any other data requiring special ethical clearance. The information collected was limited to socio-economic characteristics and farm management practices. As this study involved non-interventional socio-economic surveys and did not include medical research or health-related data, the provisions of the Declaration of Helsinki regarding medical research on human subjects were not applicable. Nevertheless, the study adhered to internationally recognized ethical standards for research involving human participants. In accordance with national regulations and standard practices for socio-economic field surveys in Côte d’Ivoire, this study did not require formal approval from an institutional ethics committee or IRB. To ensure confidentiality, all personal data were anonymized following a predefined data protection procedure. Interviews were conducted mainly in French or local languages by the first author, with the assistance of local interpreters when necessary. Data collection and survey design Data were collected using a semi-structured questionnaire administered to 158 cocoa smallholder households through a stratified random sampling approach. Stratification was based on geographic, pedoclimatic, and sociodemographic characteristics of the study areas, a method widely recognized in agronomic and rural surveys for ensuring representative coverage across distinct ecological and social strata64. To provide a representative overview of cocoa farming systems in Côte d’Ivoire, five study zones were selected: Maféré, Grand-Béréby, Divo, Yakassé-Attobrou (Biébi), and Guibéroua. Surveyed households were allocated according to field accessibility, availability of cocoa-farming households, and logistical constraints, resulting in 158 respondents distributed as follows: Maféré (18), Grand-Béréby (21), Divo (42), Yakassé-Attobrou (43), and Guibéroua (34). One respondent per household was interviewed, corresponding to the household head or, in their absence, the primary decision-maker responsible for cocoa farm management. Household enumerations were conducted through door-to-door visits, supported by local guides, in line with established best practices in rural socio-agrarian fieldwork, as exemplified in climate adaptation surveys in Ghana65. The questionnaire was developed based on a comprehensive literature review33,64 and pre-tested with ten “reference farmers” per study zone to ensure clarity, relevance, and local appropriateness. To account for potential imbalances in sample sizes across zones, statistical analyses incorporated study zone as a random effect in generalized linear mixed models (GLMMs), enabling robust comparisons across heterogeneous strata. For analytical purposes, the collected variables were grouped into three main categories: Sociodemographic characteristics of cocoa farmers, including age, education level, gender, land access mode, farmer’s experience, and origin. These variables provide insight into the social and human factors influencing farm management decisions and productivity.Plantation attributes and agronomic practices, such as plot size, planting design, input use, and annual cocoa production, reflecting management intensity and key yield determinants.Farmers’ perceptions of major biophysical constraints, including pests, diseases, and soil fertility issues, offering a farmer-centered perspective on the challenges affecting cocoa productivity and sustainability. Sociodemographic characteristics of cocoa farmers, including age, education level, gender, land access mode, farmer’s experience, and origin. These variables provide insight into the social and human factors influencing farm management decisions and productivity. Sociodemographic characteristics of cocoa farmers, including age, education level, gender, land access mode, farmer’s experience, and origin. These variables provide insight into the social and human factors influencing farm management decisions and productivity. Plantation attributes and agronomic practices, such as plot size, planting design, input use, and annual cocoa production, reflecting management intensity and key yield determinants. Plantation attributes and agronomic practices, such as plot size, planting design, input use, and annual cocoa production, reflecting management intensity and key yield determinants. Farmers’ perceptions of major biophysical constraints, including pests, diseases, and soil fertility issues, offering a farmer-centered perspective on the challenges affecting cocoa productivity and sustainability. Farmers’ perceptions of major biophysical constraints, including pests, diseases, and soil fertility issues, offering a farmer-centered perspective on the challenges affecting cocoa productivity and sustainability. Statistical analysis After data collection, responses were entered and cleaned in Microsoft Excel and then prepared for statistical analysis. All variables were converted into categorical formats, except for cocoa yield, which was retained as a continuous quantitative variable. Given the geographical and agroecological diversity of the study sites, both study area and ecological sector were included as explanatory variables to assess their potential influence on cocoa yield. Soil types were excluded from the statistical analysis because they represent dominant but not exclusive conditions within sites and were not identified at the plot level. The initial analytical phase consisted of descriptive statistics to examine the distribution of variables, including yield variation across explanatory factors and study zones. These analyses were conducted on the full dataset of 158 observations. Since several variables exhibited non-normal distributions, non-parametric tests were applied. Specifically, the Mann–Whitney U test was used to compare yield distributions between independent groups (e.g., study areas and ecological sectors), given its appropriateness when normality assumptions are not met66. In addition, Fisher’s exact test was employed to assess associations between categorical variables and study zones, particularly suitable for small or unbalanced samples67. Together, these tests provided robust insights into distributional differences and categorical associations relevant to cocoa productivity. To identify key factors associated with yield, a generalized linear mixed model (GLMM) was fitted using a Gamma distribution with a log-link function68. Because of missing yield data, the model was estimated on a reduced sample of 123 observations. The GLMM included both sociodemographic characteristics and agronomic plantation variables as fixed effects, while accounting for spatial heterogeneity by including study zone as a random effect. Variable selection followed a backward stepwise procedure based on the Akaike Information Criterion (AIC), with the subsequent reintroduction of theoretically relevant predictors to ensure model robustness69. Formally, the model was expressed as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y_{i}\sim Gamma\left( {\mu_{i},\sigma ^2} \right)$$\end{document}2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{log}\left({\mu}_{i}\right)={{\upeta}}_{i}$$\end{document} with the linear predictor:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\upeta}}_{i}={\beta}_{0}+{\beta}_{1}{X}_{1i}+{\beta}_{2}{X}_{2i}+\dots+{\beta}_{13}{X}_{13i}+{\mathrm{u}}_{j}$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y}_{i}$$\end{document} denotes the observed cocoa yield for plot i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu}_{i}$$\end{document} its expected value, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma ^{ \wedge } 2}$$\end{document} the dispersion parameter of the Gamma distribution, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\upeta}}_{i}$$\end{document} the linear predictor, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta}_{0}$$\end{document} the intercept and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta}_{k}$$\end{document} (k = 1, …, 13) the regression coefficients associated with covariates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{ki}$$\end{document}. The covariates included \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{1}$$\end{document} = Insecticide treatment frequency, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{2}$$\end{document} = Farmers age, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{3}$$\end{document} = Previous land use, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{4}$$\end{document} = Age of plantation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{5}$$\end{document} = Planting design, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{6}$$\end{document} = Planting material, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{7}$$\end{document} = Planting mode, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{8}$$\end{document} = Granular fertilizer, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{9}$$\end{document} = Farmer’s gender, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{10}$$\end{document} = Farmer’s experience, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{11}$$\end{document} = Farmer’s age, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{12}$$\end{document} = Land acquisition mode, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{13}$$\end{document} = Education level and the term \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$uj \sim N\left( {0,\tau ^{{ \wedge 2}} } \right)$$\end{document} is the random intercept for study zone j with variance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau}^{^2}$$\end{document}. On the multiplicative scale, the expected yield was given by:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu}_{i}=\mathrm{exp}\left({\beta}_{0}\right)\times\prod_{k=1}^{13}\mathrm{e}\mathrm{x}\mathrm{p}({\beta}_{k}{X}_{ki})\times\mathrm{e}\mathrm{x}\mathrm{p}({\mathrm{u}}_{j})$$\end{document} where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{ki}$$\end{document} denotes the k-th covariate for observation i. For categorical predictors, coefficients were interpreted relative to their respective reference categories (Table 4). Model validity was assessed through residual diagnostics suitable for Gamma GLMMs70. Both marginal (fixed effects only) and conditional (fixed + random effects) R² values were computed following66 to quantify explained variance. Parameter significance was evaluated using 95% confidence intervals, and statistical significance was established at p < 0.05. All statistical analyses were performed using R software (version 4.2.2) and the GLMM model was implemented using the glmmTMB package71,72. Results Sociodemographic profile of cocoa farmers The results of the survey are summarized in Table 1. A total of 158 households were surveyed across the five study sites. Most cocoa producers were male, with proportions ranging from 88.37% to 100% depending on the study area. No statistically significant differences in gender distribution were observed between sites (P = 0.32). Land acquisition methods varied significantly across zones (P < 0.001). Inheritance was the predominant mode of land access in Divo (71.43%) and Yakassé-Attobrou (Biébi) (88.37%) but was much less common in Maféré (5.56%) and Guibéroua (26.47%). Land purchase was most frequent in Grand-Béréby (57.14%) and Guibéroua (32.35%), while sharecropping dominated in Maféré (83.33%). Producers’ origins varied significantly across locations (P < 0.001). Allochthonous farmers were more numerous in Divo (42.86%) and Grand-Béréby (28.57%). Exogenous migrants predominated in Guibéroua (55.88%), Grand-Béréby (71.43%), and Maféré (100%). Autochthonous farmers represented the majority in Yakassé-Attobrou (Biébi) (86.05%) and Divo (26.19%). Educational attainment showed wide variation across sites (P < 0.001). In Maféré, most producers had no formal education (88.89%), while in Yakassé-Attobrou (Biébi), more than half had completed secondary education (58.14%). Lower proportions of producers with secondary education were recorded in Divo (28.57%), Guibéroua (20.59%), and Grand-Béréby (14.29%). Experience in cocoa cultivation also varied considerably across the zones. Most producers reported 16–30 years of experience, particularly in Divo (69.05%) and Grand-Béréby (66.67%). In Yakassé-Attobrou (Biébi), a larger proportion of producers reported 3–15 years of experience (37.21%). In terms of age distribution, most producers were between 27 and 45 years old in most study areas (53.49%–64.29%). However, in Maféré, a notable proportion of producers were older, with 55.56% aged 46–60 years. Table 1Cocoa farmer profiles in the study zonesStudy areaDivo(N=42)Grand-Béréby(N=21)Guiberoua(N=34)Maféré(N=18)Yakassé-Attobrou (Biébi)(N=43)All sites(N=158) P-valueGenderMale97.6295.2497.0610088.3796.690.32Female2.384.762.94011.633.31Land access modeInherited71.4333.3326.475.5688.3756.20.00***Sharecropping2.389.5241.1883.339.319.83Purchased26.1957.1432.3511.112.3323.97Farmers originAllochthonous42.8628.5726.4709.326.450.00***Exogenous30.9571.4355.881004.6538.02Autochthonous26.19017.65086.0535.54Education levelNone47.6266.675088.8916.2843.750.00***Primary school23.8119.0529.415.5625.5823.97Secondary school28.5714.2920.595.5658.1432.23Farmer’s experience[3; 15]23.8119.0529.4116.6737.2161.980.34[16; 30]69.0566.6758.8255.5646.5125.62[31; 45]7.1414.2911.7627.7816.2812.4Farmers age[27; 45]64.2961.955.8838.8953.4957.850.41[46; 60]33.3338.129.4155.5641.8638.84[61; 83]2.38011.765.564.653.31P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)Autochthonous: farmers native to the village or study area (local farmers), Allochthonous: farmers originating from other regions of Côte d’Ivoire (internal migrants), Exogenous: farmers originating from other countries (foreign migrants).Sharecropping: A farming arrangement in which the farmer cultivates a plot owned by a landowner. Once the cocoa trees reach maturity, the plot is divided between the landowner and the farmer, with the latter gaining ownership of their portion. Table 1Cocoa farmer profiles in the study zonesStudy areaDivo(N=42)Grand-Béréby(N=21)Guiberoua(N=34)Maféré(N=18)Yakassé-Attobrou (Biébi)(N=43)All sites(N=158) P-valueGenderMale97.6295.2497.0610088.3796.690.32Female2.384.762.94011.633.31Land access modeInherited71.4333.3326.475.5688.3756.20.00***Sharecropping2.389.5241.1883.339.319.83Purchased26.1957.1432.3511.112.3323.97Farmers originAllochthonous42.8628.5726.4709.326.450.00***Exogenous30.9571.4355.881004.6538.02Autochthonous26.19017.65086.0535.54Education levelNone47.6266.675088.8916.2843.750.00***Primary school23.8119.0529.415.5625.5823.97Secondary school28.5714.2920.595.5658.1432.23Farmer’s experience[3; 15]23.8119.0529.4116.6737.2161.980.34[16; 30]69.0566.6758.8255.5646.5125.62[31; 45]7.1414.2911.7627.7816.2812.4Farmers age[27; 45]64.2961.955.8838.8953.4957.850.41[46; 60]33.3338.129.4155.5641.8638.84[61; 83]2.38011.765.564.653.31P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)Autochthonous: farmers native to the village or study area (local farmers), Allochthonous: farmers originating from other regions of Côte d’Ivoire (internal migrants), Exogenous: farmers originating from other countries (foreign migrants).Sharecropping: A farming arrangement in which the farmer cultivates a plot owned by a landowner. Once the cocoa trees reach maturity, the plot is divided between the landowner and the farmer, with the latter gaining ownership of their portion. Cocoa farmer profiles in the study zones P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)Autochthonous: farmers native to the village or study area (local farmers), Allochthonous: farmers originating from other regions of Côte d’Ivoire (internal migrants), Exogenous: farmers originating from other countries (foreign migrants).Sharecropping: A farming arrangement in which the farmer cultivates a plot owned by a landowner. Once the cocoa trees reach maturity, the plot is divided between the landowner and the farmer, with the latter gaining ownership of their portion. Farm characteristics and agricultural practices Agricultural practices differed significantly across the study zones. Insect pest management practices differed markedly between zones (P < 0.001). In Yakassé-Attobrou, most producers (65.12%) applied few or no insecticide treatments (0–3 per year), whereas intensive spraying (6–12 applications annually) was predominant in Grand-Béréby (80.95%). Guibéroua and Maféré showed intermediate, with most farmers applying 4–5 treatments per year. Divo exhibited a relatively balanced distribution across categories (Table 2). Table 2Farm management practices by farmers (%) across the study area.Study areaDivo(N = 42)Grand-Béréby(N = 21)Guiberoua(N = 34)Maféré(N = 18)Yakassé-Attobrou (Biébi)(N = 43)All sites(N = 158)P-valueInsecticide treatment frequency[0; 3]33.339.5238.245.5665.1233.880.00***[4;5]45.249.5258.8272.2234.8846.28[6; 12]21.4380.952.9411.11019.83Granular fertilizerNo80.9547.6264.7127.7867.4461.160.00***Yes19.0552.3835.2972.2232.5638.84Planting designNot aligned85.7195.2470.5910095.3588.430.00***aligned14.294.7629.4104.6511.57Planting materialKnown23.8128.5738.2422.2227.9127.270.69Unknown76.1971.4361.7677.7872.0972.73Planting modeDirect seeding14.2933.3323.5372.2288.3744.630.00***Nursery69.0542.8661.765.56042.15Seeding and nursery16.6723.8114.7122.2211.6313.22Age of plantation[1.5; 15]45.2423.8141.1827.7823.2650.410.17[16; 30]40.4852.3844.1266.6760.4732.23> 3014.2923.8114.71016.2817.36Previous land useFallow54.7657.1426.4711.119.334.710.00***Forest45.2442.8673.5388.8990.765.29Farm size[0.5; 2]61.961.95066.6751.1655.370.03**]2.5; 5]33.3328.575016.6739.5338.84]5.5; 10.5]4.769.52011.116.985.79P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing) Table 2Farm management practices by farmers (%) across the study area.Study areaDivo(N = 42)Grand-Béréby(N = 21)Guiberoua(N = 34)Maféré(N = 18)Yakassé-Attobrou (Biébi)(N = 43)All sites(N = 158)P-valueInsecticide treatment frequency[0; 3]33.339.5238.245.5665.1233.880.00***[4;5]45.249.5258.8272.2234.8846.28[6; 12]21.4380.952.9411.11019.83Granular fertilizerNo80.9547.6264.7127.7867.4461.160.00***Yes19.0552.3835.2972.2232.5638.84Planting designNot aligned85.7195.2470.5910095.3588.430.00***aligned14.294.7629.4104.6511.57Planting materialKnown23.8128.5738.2422.2227.9127.270.69Unknown76.1971.4361.7677.7872.0972.73Planting modeDirect seeding14.2933.3323.5372.2288.3744.630.00***Nursery69.0542.8661.765.56042.15Seeding and nursery16.6723.8114.7122.2211.6313.22Age of plantation[1.5; 15]45.2423.8141.1827.7823.2650.410.17[16; 30]40.4852.3844.1266.6760.4732.23> 3014.2923.8114.71016.2817.36Previous land useFallow54.7657.1426.4711.119.334.710.00***Forest45.2442.8673.5388.8990.765.29Farm size[0.5; 2]61.961.95066.6751.1655.370.03**]2.5; 5]33.3328.575016.6739.5338.84]5.5; 10.5]4.769.52011.116.985.79P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing) Farm management practices by farmers (%) across the study area. P-values were obtained using Fisher’s Exact Test to assess the association between variables and study areas. Values below 0.05 indicate statistically significant differences across study areas.Significance levels between variables: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing) The use of mineral fertilizers also varied significantly across zones (P < 0.001). Adoption rates were highest in Maféré (72.22%) and Grand-Béréby (52.38%), but considerably lower in Divo (19.05%) and Yakassé-Attobrou (32.56%). Overall, 38.84% of producers reported applying mineral fertilizers (Table 2). Proper cocoa tree alignment was generally uncommon, with only 11.57% of plantations displaying appropriate spacing (P < 0.001). Guibéroua was an exception (29.41%), while Maféré showed universal irregular planting (100%) (Table 2). Regarding planting material, uncertified, locally sourced seedlings were predominant (72.73%). Direct seeding was prevalent in Yakassé-Attobrou (88.37%) and Maféré (72.22%), whereas nursery-raised seedlings were more common in Divo (69.05%) and Guibéroua (61.76%). The mixed use of both methods was relatively rare (13.22%). No statistically significant differences were observed in plantation age across sites (P = 0.17). Relatively young plantations (1.5–15 years) were more common in Divo (45.24%), whereas medium-aged plantations (16–30 years) predominated in Maféré (66.67%) and Yakassé-Attobrou (60.47%). Only a small proportion of plantations were older than 30 years (Table 2). Patterns of land use before cocoa establishment showed significant variation across sites (P < 0.001). In Maféré and Yakassé-Attobrou, plantations were mainly established on cleared forest land (88.89% and 90.7%, respectively), whereas in Divo (54.76%) and Grand-Béréby (57.14%), they were more commonly established on previously fallow land (Table 2). Finally, farm size differed significantly across zones (P = 0.03). Smallholdings (0.5–2 ha) were predominant in Maféré (66.67%) and Divo (61.9%), whereas medium to large farms (> 2 ha) were more frequent in Guibéroua and Grand-Béréby. Very large plantations (> 5 ha) were uncommon across all sites (Table 2) (Fig. 1). Spatial variability of yield As shown in Table 3; Fig. 2, mean cocoa yields varied across the different study zones and ecological sectors (Fig. 3). Table 3Cocoa yield across the study areas and ecological sector.VariablesCategoryAverage cocoa yield(kg ha− 1 year− 1 ± standard deviation)P valueStudy areasGrand-Béréby812.13 ± 381.92 a0.01**Guiberoua790.495 ± 469.69 abDivo481.94 ± 259.12 bcMaféré480.3 ± 283.29 abcYakassé-Attobrou (Biébi)372.07 ± 196.49 cEcological sectorsOmbrophilic sector512.25 ± 330.03 a0.2433semideciduous zone588.34 ± 372.84 a Table 3Cocoa yield across the study areas and ecological sector.VariablesCategoryAverage cocoa yield(kg ha− 1 year− 1 ± standard deviation)P valueStudy areasGrand-Béréby812.13 ± 381.92 a0.01**Guiberoua790.495 ± 469.69 abDivo481.94 ± 259.12 bcMaféré480.3 ± 283.29 abcYakassé-Attobrou (Biébi)372.07 ± 196.49 cEcological sectorsOmbrophilic sector512.25 ± 330.03 a0.2433semideciduous zone588.34 ± 372.84 a Cocoa yield across the study areas and ecological sector. Fig. 1Geographical location of the study areas. Geographical location of the study areas. Fig. 2Boxplots showing yield variation across study areas (Kruskal-Wallis test) and ecological sectors (Mann-Whitney test). Boxplots showing yield variation across study areas (Kruskal-Wallis test) and ecological sectors (Mann-Whitney test). Fig. 3Socio-demographic factors and agricultural practices influencing cocoa yield. Socio-demographic factors and agricultural practices influencing cocoa yield. Cocoa yield variability across study zones Significant differences in average cocoa yields were observed among the study zones. The highest yields were recorded in Grand-Béréby (812.13 kg ha⁻¹ year⁻¹) and Guibéroua (790.49 kg ha⁻¹ year⁻¹), whereas Yakassé-Attobrou (Biébi) exhibited substantially lower yields (372.07 kg ha⁻¹ year⁻¹). Divo and Maféré recorded intermediate and nearly identical yields, at 481.94 kg ha⁻¹ year⁻¹ and 480.30 kg ha⁻¹ year⁻¹, respectively (Table 3; Fig. 2). Influence of ecological sectors on cocoa yield Analysis by ecological sector indicated that the mesophilic zone (588.34 kg ha⁻¹ year⁻¹) had a higher mean cocoa yield than the ombrophilous zone (512.25 kg ha⁻¹ year⁻¹); however, this difference was not statistically significant (P = 0.2433) (Table 3; Fig. 2). Socio-demographic and management practices influencing cocoa yield The Generalized Linear Mixed Model (GLMM), accounting for study zones as random effects, identified several significant determinants of cocoa yield (Table 4; Fig. 3). Fig. 4Key biophysical constraints across all study areas (Fisher’s Exact Test). Key biophysical constraints across all study areas (Fisher’s Exact Test). Among agronomic practices, cocoa tree alignment, type of planting material, and mineral fertilizer use were significant. Plots with cocoa trees planted in aligned rows showed higher yields than those with irregular planting (Estimate = 0.41; P = 0.009). The use of selected planting material or productive clones was positively associated with yield (Estimate = 0.21; P = 0.049), as was the application of mineral fertilizer (Estimate = 0.25; P = 0.0365). Regarding sociodemographic characteristics, farmer age was significantly associated with yield. Producers aged 61–83 years recorded lower yields than younger farmers aged 27–45 years (Estimate = − 0.73; P = 0.0021), with a clear decreasing trend across age groups. Secondary education level showed a marginally negative association with yield (Estimate = −0.256; P = 0.0784). Other variables—including frequency of insecticide treatments, cultivated area, land tenure mode, gender, and planting layout—were not significant in this multivariate model (Table 4; Fig. 4). Table 4Main agronomic and socioeconomic determinants of cocoa yield in smallholder Variables (modalities)EstimateStd. errorz valuePr(>|z|)(Intercept)6.0720.35517.093< 2e-16***Agronomic management practicesPesticide treatment frequency ([3 ;5])0.160.1111.4420.149Pesticide treatment frequency ([6 ;12])0.0960.1820.530.596Farmers age ([2.5; 5])-0.0410.102-0.4050.685Farmers age ([5.5; 10.5])-0.1540.209-0.7370.461Previous land use (Fallow)-0.0290.111-0.2640.792Age of plantation ([16 ;30])-0.0840.122-0.6880.492Age of plantation (>30)-0.1370.165-0.8290.407Planting design (aligned)0.410.1582.5870.01*Planting material (Known)0.2110.1071.9620.004**Planting mode (Nursery)0.0310.130.2430.808Planting mode (Seeding and nursery)0.0560.1570.3560.722Granular fertilizer (Yes)0.2540.1222.0920.036*Socio-demographic profile characteristicsGender (Male)0.090.2830.3180.751Farmer experience ([16; 30])0.0640.1140.5610.575Farmer experience ([31; 45])0.2270.1881.2060.228Farmers age ([46; 60])-0.2170.119-1.8280.068†Farmers age ([61; 83])-0.730.237-3.0760.002**Land access mode (sharecropping)-0.0820.181-0.4530.65Land access mode (Purchased)-0.0880.147-0.6010.548Education level (Primary school)0.0140.1270.1070.915Education level (Secondary school)-0.2560.146-1.760.078†Residual deviance = 1687.8; AIC = 1735.8; BIC = 1803.3; Dispersion estimates for Gamma family (sigma^2) = 0.225; Random effects: Study area. Random effect variance = 0.05058. Generalized Linear; Marginal R² (fixed effect) = 0.238. Conditional R² (total) = 0.378Mixed Model (GLMM): P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1Reference categories are: [0;2] for Pesticide treatment frequency. [0.5;2] for Farmers age. “Forest” for Previous land use. [1.5 ;15] for Age of plantation. “Not aligned” for Planting design. “Unknown” for Planting material. “Direct seeding” for Planting mode. “No” for Granular fertilizer. “Female” for Gender. [16;30] for Farmer experience. [27;45] for Farmers age. “Inherited” for Land access mode. and “None” for Education level. Table 4Main agronomic and socioeconomic determinants of cocoa yield in smallholder Variables (modalities)EstimateStd. errorz valuePr(>|z|)(Intercept)6.0720.35517.093< 2e-16***Agronomic management practicesPesticide treatment frequency ([3 ;5])0.160.1111.4420.149Pesticide treatment frequency ([6 ;12])0.0960.1820.530.596Farmers age ([2.5; 5])-0.0410.102-0.4050.685Farmers age ([5.5; 10.5])-0.1540.209-0.7370.461Previous land use (Fallow)-0.0290.111-0.2640.792Age of plantation ([16 ;30])-0.0840.122-0.6880.492Age of plantation (>30)-0.1370.165-0.8290.407Planting design (aligned)0.410.1582.5870.01*Planting material (Known)0.2110.1071.9620.004**Planting mode (Nursery)0.0310.130.2430.808Planting mode (Seeding and nursery)0.0560.1570.3560.722Granular fertilizer (Yes)0.2540.1222.0920.036*Socio-demographic profile characteristicsGender (Male)0.090.2830.3180.751Farmer experience ([16; 30])0.0640.1140.5610.575Farmer experience ([31; 45])0.2270.1881.2060.228Farmers age ([46; 60])-0.2170.119-1.8280.068†Farmers age ([61; 83])-0.730.237-3.0760.002**Land access mode (sharecropping)-0.0820.181-0.4530.65Land access mode (Purchased)-0.0880.147-0.6010.548Education level (Primary school)0.0140.1270.1070.915Education level (Secondary school)-0.2560.146-1.760.078†Residual deviance = 1687.8; AIC = 1735.8; BIC = 1803.3; Dispersion estimates for Gamma family (sigma^2) = 0.225; Random effects: Study area. Random effect variance = 0.05058. Generalized Linear; Marginal R² (fixed effect) = 0.238. Conditional R² (total) = 0.378Mixed Model (GLMM): P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1Reference categories are: [0;2] for Pesticide treatment frequency. [0.5;2] for Farmers age. “Forest” for Previous land use. [1.5 ;15] for Age of plantation. “Not aligned” for Planting design. “Unknown” for Planting material. “Direct seeding” for Planting mode. “No” for Granular fertilizer. “Female” for Gender. [16;30] for Farmer experience. [27;45] for Farmers age. “Inherited” for Land access mode. and “None” for Education level. Main agronomic and socioeconomic determinants of cocoa yield in smallholder Residual deviance = 1687.8; AIC = 1735.8; BIC = 1803.3; Dispersion estimates for Gamma family (sigma^2) = 0.225; Random effects: Study area. Random effect variance = 0.05058. Generalized Linear; Marginal R² (fixed effect) = 0.238. Conditional R² (total) = 0.378Mixed Model (GLMM): P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1Reference categories are: [0;2] for Pesticide treatment frequency. [0.5;2] for Farmers age. “Forest” for Previous land use. [1.5 ;15] for Age of plantation. “Not aligned” for Planting design. “Unknown” for Planting material. “Direct seeding” for Planting mode. “No” for Granular fertilizer. “Female” for Gender. [16;30] for Farmer experience. [27;45] for Farmers age. “Inherited” for Land access mode. and “None” for Education level. The model explained 23.8% of the variance in yield through fixed effects alone (marginal R²) and 37.8% when including random effects of study zones (conditional R²). Biophysical constraints of cocoa farming according to farmers across study area The analysis of farmers’ perceptions regarding biophysical constraints to cocoa production revealed considerable variability across the study sites (Table 5; Fig. 4). Table 5Producers’ perceptions of biophysical constraints in cocoa production across study sitesDivo (N=42)Grand-Béréby (N=21)Guibéroua (N=34)Maféré (N=18)Yakassé-Attobrou (Biebi) (N=43)All sites(N=158) P-value(constraint × study area)Pest damages52.3871.435033.3325.5844.940.005**Brown pod rot11.938.18.8210079.0743.04< 0.001***Low soil fertility38.152.3832.3538.899.331.010.002**Mortality of cocoa trees14.2928.5732.3527.7834.8827.220.224Drought30.959.5229.41018.620.890.019*Loranthus sp.16.6728.5720.59023.2618.990.128Green moss014.292.9411.119.36.330.062 †Weeds2.3802.945.5611.635.060.284Termite damage7.1402.9406.984.430.702Plantation aging4.764.76004.653.160.653Significance levels between study sites: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)The p-values were obtained using the Kruskal-Wallis or Mann-Whitney test to compare farmers’ perceptions of biophysical constraints across site. Table 5Producers’ perceptions of biophysical constraints in cocoa production across study sitesDivo (N=42)Grand-Béréby (N=21)Guibéroua (N=34)Maféré (N=18)Yakassé-Attobrou (Biebi) (N=43)All sites(N=158) P-value(constraint × study area)Pest damages52.3871.435033.3325.5844.940.005**Brown pod rot11.938.18.8210079.0743.04< 0.001***Low soil fertility38.152.3832.3538.899.331.010.002**Mortality of cocoa trees14.2928.5732.3527.7834.8827.220.224Drought30.959.5229.41018.620.890.019*Loranthus sp.16.6728.5720.59023.2618.990.128Green moss014.292.9411.119.36.330.062 †Weeds2.3802.945.5611.635.060.284Termite damage7.1402.9406.984.430.702Plantation aging4.764.76004.653.160.653Significance levels between study sites: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)The p-values were obtained using the Kruskal-Wallis or Mann-Whitney test to compare farmers’ perceptions of biophysical constraints across site. Producers’ perceptions of biophysical constraints in cocoa production across study sites Significance levels between study sites: P < 0.001 (***); 0.001 ≤ P < 0.01 (**); 0.001 ≤ P < 0.01 (**); 0.05 ≤ P < 0.1 (†); P ≥ 0.1 (Nothing)The p-values were obtained using the Kruskal-Wallis or Mann-Whitney test to compare farmers’ perceptions of biophysical constraints across site. Pest attacks were the most frequently reported constraint overall, cited by 44.94% of farmers, with significant differences among zones (P = 0.005). The prevalence of pest damage decreased in the following order: Grand-Béréby (71.43%), Divo (52.38%), Guibéroua (50%), Maféré (33.33%), and Yakassé-Attobrou (25.58%). Brown pod rot (Phytophthora spp.) was the second most frequently reported constraint, exhibiting highly significant spatial variation (P < 0.001). It was particularly prevalent in Maféré (100%), Yakassé-Attobrou (79.07%), and Grand-Béréby (38.1%), whereas its incidence was low in Divo (11.9%) and Guibéroua (8.82%). Low soil fertility ranked third, reported by 31.01% of farmers, with significant differences among zones (P = 0.002). In Grand-Béréby, more than half of the producers (52.38%) identified limited soil fertility as a major constraint. Maféré (38.89%), Divo (38.1%), and Yakassé-Attobrou (32.35%) were close to the overall average, whereas only 9.3% of producers in Guibéroua cited this constraint. Cocoa tree mortality was reported by 27.22% of farmers, with frequencies ranging from 14.29% to 34.88% depending on the locality; however, differences among zones were not statistically significant. Drought affected 20.89% of producers, with higher incidence in Divo (30.95%), Guibéroua (29.41%), and Yakassé-Attobrou (18.6%), but it was virtually absent in Maféré (0%) and Grand-Béréby (9.52%), showing significant variation across zones (P = 0.019). Other constraints—including Loranthus sp. and green mosses on tree trunks, weed infestation, termite damage, and plantation aging—were reported at lower frequencies (3–18%). Among these, only the prevalence of green mosses approached statistical significance across zones (P = 0.062). Overall, these results indicate that farmers’ perceptions of biophysical constraints vary considerably across study areas, likely reflecting differences in local agroecological conditions and phytosanitary dynamics specific to each region. Discussion Sociodemographic structure and land tenure dynamics The findings of this study confirm the strong dominance of male landowners in the cocoa sector, with men representing over 88% of the surveyed population across all regions. Such gender disparity has been widely reported in previous studies19,30,33,73–75 and is often attributed to the labor-intensive nature of cocoa farming combined with traditional sociocultural norms that shape land inheritance and labor roles76,77. However, these figures may overshadow the meaningful, yet often invisible, contributions of women, particularly in pre- and post-harvest activities within the cocoa value chain76,78. This marginalization raises critical concerns about equity and access to productive resources, underscoring the need for gender-sensitive policies and interventions that recognize women’s indispensable roles73,78,79. Distinct patterns of land acquisition emerged across study sites, reflecting varied customary tenure systems. In autochthonous-dominated areas such as Yakassé-Attobrou and Divo, land transfer occurs predominantly through inheritance, highlighting the enduring importance of lineage-based rights46,80. Conversely, in migrant-dense zones like Grand-Béréby, Guibéroua, Maféré, and Biankouma, land is more often acquired through purchases and sharecropping. For instance, in Biankouma, previous studies46 indicate that migrants primarily access land via these mechanisms due to limited inheritance opportunities. Educational disparities among cocoa farmers are notable. A high proportion of producers in Maféré and Grand-Béréby reported having received no formal schooling, likely reflecting the substantial presence of migrants from Mali and Burkina Faso. In contrast, over half of the producers in Yakassé-Attobrou (58.14%) attained secondary education, suggesting a more structured socio-educational environment and greater capacity to adopt innovations. Nevertheless, despite this higher educational level, yields in Yakassé-Attobrou remain lower than those in Maféré and Grand-Béréby. This highlights that education, while important for productivity81–83, is insufficient alone; its benefits are realized only when combined with effective adoption of improved agricultural practices84. Farmers’ experience levels predominantly ranged from 16 to 30 years, emphasizing the value of a mature and skilled workforce for sustainable cocoa production. Previous research85 has shown a positive correlation between farmers’ experience and their likelihood of adopting pest and disease management practices, as well as other advanced agricultural technologies. However, the presence of less-experienced yet well-educated farmers in Yakassé-Attobrou suggests an influx of new entrants, likely motivated by educated youth turning to agriculture in the face of limited formal employment opportunities86. This relative inexperience may partially explain the lower use of pesticides and, consequently, the reduced yields observed in this area. Analysis of farmers’ age distribution further illustrates a predominant 27–45-year group, indicative of a relatively young and potentially adaptive farming population. This trend may reflect intergenerational land transfers driven by increasing land scarcity. Notably, Maféré’s older age profile (46–60 years) corresponds with the prevalence of sharecropping arrangements, suggesting that indigenous landowners may be shifting land-use priorities toward more lucrative crops, thereby limiting tenants’ opportunities for diversification86. Farm characteristics and agricultural practices Beyond sociodemographic characteristics, structural and technical attributes of cocoa farms play a pivotal role in determining productivity and sustainability87–90. Considerable heterogeneity was observed across the study zones, likely shaped by agroecological conditions, farmers’ backgrounds, and differential access to inputs and extension services—all of which influence farm efficiency and resilience9. A stark contrast emerged in pest management intensity. Farmers in Grand-Béréby and Maféré applied more insecticides, with Grand-Béréby using up to 12 treatments annually. This reflects both higher pest pressure, as reported in Sect. 3.5 and greater financial capacity, likely contributing to the higher yields recorded in these areas. In contrast, many farmers in Yakassé-Attobrou reported little to no pesticide use, leaving crops vulnerable to significant yield losses. Similarly, the adoption of granular fertilizers was significantly higher in Maféré and Grand-Béréby. This pattern may relate to pedoclimatic conditions—particularly the prevalence of Xanthic Ferralsols, which, while physically favorable, are prone to nutrient leaching under high rainfall, necessitating more frequent fertilization60. Despite agronomic recommendations advocating for proper tree alignment and regular spacing to optimize management and yield91, adoption remains limited (11.57%), compromising planting density and potential productivity. Moreover, most cocoa trees originated from unselected sources, highlighting limited access to improved planting material and underscoring the urgency of intensified varietal improvement programs to enhance yield and stress resilience. Farm sizes predominantly ranged from 0.5 to 2 hectares, consistent with national patterns observed near the Taï National Park92. While smallholdings may limit economies of scale, they can support more intensive and sustainable management if adequately supported. Collectively, these findings emphasize the need for tailored interventions that integrate both biophysical and socioeconomic factors to strengthen the resilience and productivity of cocoa systems. Spatial variability of cocoa yields Cocoa yields varied significantly across zones, reflecting the influence of both environmental and management-related factors. At the zonal level, Grand-Béréby (812.13 kg ha⁻¹ year⁻¹) and Guibéroua (790.49 kg ha⁻¹ year⁻¹) recorded the highest yields, whereas Yakassé-Attobrou exhibited the lowest (372.07 kg ha⁻¹ year⁻¹). These spatial disparities underscore the importance of agroecological heterogeneity in shaping cocoa productivity. Interestingly, when yields were analyzed by ecological sector, no significant difference (P = 0.2433) was observed between the humid ombrophilous zones (512.25 ± 330.03 kg ha⁻¹ year⁻¹) and the mesophilic transitional zones (588.34 ± 372.84 kg ha⁻¹ year⁻¹). This suggests that zonal yield differences may be driven more strongly by localized factors—such as soil fertility, planting density, input availability, and farmer experience—than by broad climatic classifications alone. In other words, cropping practices appear broadly similar among farmers in Côte d’Ivoire. For example, low productivity in Yakassé-Attobrou may stem from limited farming experience, inadequate technical support, and suboptimal resource use, whereas higher yields in Grand-Béréby and Guibéroua likely reflect more intensive input application and management93. These findings align with studies from Togo, where no significant differences in cocoa yields were observed across ecological zones, suggesting that institutional and socio-cultural factors may modulate the influence of ecological gradients55. In contrast, research from Ghana reported yields ranging from 288 kg ha⁻¹ year⁻¹ in drier areas to over 849 kg ha⁻¹ year⁻¹ in more humid zones, highlighting the critical role of water availability in shaping cocoa productivity94. Collectively, these results underscored the need for context-specific strategies that integrate both biophysical and socio-economic dimensions to sustainably enhance cocoa yields. Determinants of cocoa yield The findings of this study underscore the multifactorial and context-dependent nature of cocoa yield, reflecting the combined influence of technical practices and socio-demographic attributes. This complexity has been widely documented across West Africa, where human and management factors are central to the performance of cocoa-based systems23,40,46,95. In this analysis, a Gamma Generalized Linear Mixed Model (GLMM) with a log link and site-level random effects allowed robust estimation of key yield determinants while accounting for ecological and geographic heterogeneity. Among technical variables, cocoa tree alignment emerged as a particularly significant predictor of yield. Plots with irregular or non-linear planting layouts consistently exhibited lower outputs. Quantitatively, aligned plots produced on average 0.41 units more than non-aligned plots, all else being equal. In multiplicative terms, this corresponds to an expected increase in yield of approximately 51% relative to the reference, highlighting the substantial productivity benefits of systematic spatial organization. These results align with96, who demonstrated that proper tree arrangement improves light penetration, air circulation, facilitates phytosanitary interventions, and reduces disease pressure—all critical for optimizing productivity. Improved light and air flow may also reduce trunk moisture and limit green moss growth, further supporting tree health and yield. The genetic origin of planting material was another key determinant. Plots established with known or improved material produced on average 0.21 units more than plots using unknown material, which corresponds to an approximate 23% expected yield increase. Clonal planting ensures genetic uniformity, higher tolerance to biotic stressors, and earlier fruiting, supporting findings by97–99 that the diffusion of improved clones is pivotal in narrowing yield gaps, especially when combined with optimized management practices. Fertilization practices, particularly the application of granular mineral fertilizers, were also positively associated with yield. Fertilized plots gained on average 0.25 units relative to unfertilized plots, corresponding to an expected 28% increase, all else being equal. This underscores the importance of nutrient inputs in promoting tree growth, flowering, and pod development, while highlighting the need for improved access to fertilizers and extension support9,100,101. From a socio-demographic perspective, advanced farmer age (61–83 years) was strongly and negatively associated with cocoa yield, corresponding to an expected 52% decrease relative to the reference category. This decline likely reflects a combination of reduced physical capacity, limited access to labor, and lower adoption of labor-intensive and innovation-driven practices, such as regular pesticide application, pruning, and replanting102. Farmers aged 46–60 also exhibited a moderate yield reduction (approximately 20%), suggesting a progressive vulnerability linked to aging rather than a sudden productivity threshold. These findings indicate that older producers constitute a structurally vulnerable group within cocoa systems, particularly in contexts where mechanization is minimal and external labor is costly or scarce. Interestingly, secondary education was marginally negatively associated with yield, likely because farmers with general secondary education often lack specialized agricultural training, have limited hands-on experience, and are less accustomed to the physically demanding tasks characteristic of smallholder cocoa farming. Consequently, they may adopt intensive management practices more slowly and diversify into non-agricultural income activities, which can reduce overall farm productivity. Together, these estimates represent conditional mean effects in a log-linked GLMM, meaning each factor affects yield independently once control variables are accounted for. Expressed on the original scale, the coefficients reflect multiplicative changes in expected yield. This integrated quantitative and qualitative interpretation highlights how technical practices and socio-demographic attributes jointly shape cocoa productivity. The explanatory power of the GLMM was moderate, with fixed effects accounting for 23.8% of yield variance and inclusion of random effects raising this to 37.8%, emphasizing the importance of both management interventions and localized environmental conditions. The findings point to the need for a holistic, context-sensitive approach to improving cocoa yields. Agronomic interventions—such as proper tree alignment, deployment of improved planting material, and targeted fertilization—must be combined with broader socio-economic strategies that address structural constraints, incentivize innovation adoption, and support vulnerable groups, particularly older producers. Enhancing access to quality inputs, strengthening extension services, and promoting youth involvement are strategic priorities for ensuring the sustainability and resilience of cocoa production systems. Cocoa farmers’ perceptions of biophysical constraints The analysis of cocoa farmers’ perceptions regarding biophysical constraints reveals pronounced spatial differentiation in both the nature and intensity of limiting factors. Pest pressure emerged as the most frequently reported constraint, cited by 44.9% of producers, with particularly high incidence in Grand-Béréby (71.4%). This aligns with findings from Méagui, where Sahlbergella singularis, Distantiella theobromae, Helopeltis sp., and Bathycoelia thalassina were identified as major Heteroptera pests exhibiting clear spatio-temporal abundance patterns103. Similarly, in the Haut-Sassandra region, mirids, stem borers, and green bugs were reported as major concerns, with stem borers notably prevalent in older orchards19. These converging observations underscore the widespread and dynamic nature of pest threats, reinforcing the need for localized and adaptive pest management strategies. Black pod rot, caused by Phytophthora spp., constitutes another major phytosanitary constraint. Its widespread incidence in Maféré and high prevalence in Yakassé-Attobrou (79%) reflect patterns previously documented by104, who identified P. palmivora as the predominant species in Côte d’Ivoire, alongside the emerging presence of P. megakarya in eastern cocoa-growing regions. The pathogen’s aggressiveness is exacerbated in humid areas with dense canopy cover105,106, while poor cultural practices—such as inadequate pruning and failure to remove infected pods—further accelerate its spread106,107. Encouragingly, several clonal varieties exhibit promising levels of resistance108, offering potential for integrated disease management strategies. Nonetheless, smallholder farmers remain highly vulnerable, as effective and accessible solutions are still lacking. Soil fertility decline, reported by 31% of farmers, was particularly pronounced in Grand-Béréby (52%) and Maféré (39%), likely reflecting nutrient depletion under continuous cocoa cultivation without adequate replenishment9. The Xanthic Ferralsols prevalent in these zones are characterized by low cation exchange capacity and high susceptibility to leaching61,109, intensifying fertility losses. Conversely, Acrisol-dominated zones such as Divo, Guibéroua, and Yakassé-Attobrou retain nutrients more effectively, corresponding to lower reports of fertility decline. Research confirms that Ferralsols and Acrisols respond differently to identical fertilizer regimes110, highlighting the need for tailored nutrient management. Integrated Soil Fertility Management (ISFM)—combining organic amendments, mineral fertilizers, improved germplasm, and context-specific practices—has proven effective for sustaining crop productivity in diverse tropical soils111. Drought stress, cited by 20.9% of respondents, was mainly reported in mesophilic zones such as Divo, Guibéroua, and Yakassé-Attobrou, characterized by irregular rainfall and prolonged dry spells29. These conditions negatively affect flowering, pod set, and tree survival. In contrast, ombrophilous areas such as Maféré and Grand-Béréby experience more consistent rainfall, rendering drought a minor concern. Agroforestry strategies incorporating drought-tolerant species have shown promise in mitigating dry-season impacts and enhancing soil moisture retention113. Although less frequently mentioned, mistletoe (Loranthus spp.), mosses, weeds, termites, and orchard aging represent context-specific threats to cocoa agroecosystems. Mistletoes, particularly hemiparasitic Loranthaceae species, can spread extensively within cocoa agroforestry systems; infestation rates exceeding 40% were reported in the Daloa region, with strong correlations between infections on shade and cocoa trees114. Moss proliferation, often associated with dense canopy and high humidity, was marginally significant (P = 0.062) and may indicate suboptimal canopy management. Termites, while ecologically beneficial for nutrient cycling, become detrimental when attacking aging or stressed trees115. Aging orchards also facilitate weed encroachment and increase susceptibility to abiotic stress, with productivity declines commonly observed after 25 years10. However, targeted orchard rehabilitation has demonstrated yield increases exceeding 200%, with economic returns surpassing 377%44, highlighting the critical need for renewal strategies in aging cocoa systems. Study limitations This study provides valuable insights into cocoa yield variability in Côte d’Ivoire; however, several limitations should be acknowledged. The analysis was restricted to five production zones and a sample of 158 households, which may limit the generalizability of the findings at the national scale. Sample sizes were not strictly equal across zones due to differences in farmer availability and accessibility. To account for this, study zone was included as a random effect in the Gamma Generalized Linear Mixed Model (GLMM) with a log link function, which is appropriate for the data and allows for robust estimates despite unequal and relatively small sample sizes. Data on farm management practices and production constraints were primarily self-reported, which may introduce recall and perception biases. In addition, the cross-sectional design provides only a temporal snapshot, preventing assessment of seasonal or long-term trends. Some potentially influential factors, such as detailed genetic characteristics of planting material, microclimatic variability, and post-harvest practices, were not included. The statistical model explained less than 40% of the observed yield variability, indicating that additional biophysical, socio-economic, or institutional determinants remain unaccounted for. Future research should aim to increase geographic coverage, expand sample sizes, and integrate longitudinal, biophysical, and farm-level management data. Developing separate models for each study zone—where sufficient sample sizes are available—would allow identification not only of general determinants but also of local, zone-specific factors influencing cocoa yield. Such improvements would enhance the robustness, representativeness, and applicability of the findings and provide more actionable insights for sustainable cocoa production. Conclusions This study aimed to identify and analyze the key factors driving cocoa yield variability across five major production zones in Côte d’Ivoire, using a multidisciplinary approach integrating social, agronomic, and ecological dimensions. The findings demonstrate that cocoa productivity is shaped by a complex interplay of farmers’ sociodemographic characteristics, farm management practices, and local biophysical conditions. The sector remains predominantly male-driven, with pronounced gender disparities in land access, decision-making, and resource allocation. Educational attainment varied across zones, often reflecting the origin of producers; however, higher education did not consistently translate into improved yields in the absence of adequate technical support and institutional infrastructure. Older farmers were generally associated with lower yields and greater vulnerability, highlighting the importance of promoting generational renewal while providing targeted support to enable older farmers to adopt improved agronomic practices. Such support may encompass tailored technical training, facilitated access to high-quality inputs, financial assistance, and advisory services. From an agronomic perspective, the use of selected planting material, systematic tree alignment, and application of mineral fertilizers emerged as key determinants of yield. Despite their positive impact, adoption of these practices remains limited due to barriers related to input availability, technical assistance, and financial capacity. Spatial analysis revealed no statistically significant differences in cocoa yields between the two ecological sectors, suggesting that localized management and socio-economic factors may exert a stronger influence on productivity than broad climatic classifications. Farmers reported insect pests, black pod disease, and declining soil fertility as major constraints, emphasizing the need for locally adapted interventions. Integrated pest and disease management strategies, alongside context-specific soil fertility management, are likely critical for sustaining cocoa productivity. Based on these findings, we recommend a set of integrated, context-sensitive policy interventions: Enhanced access to inputs and services: Improve the availability of quality planting material, fertilizers, credit, and agricultural advisory services, particularly for smallholders and older farmers who face barriers to adoption.Inclusive land policies and gender equity: Promote policies recognizing and supporting women’s contributions to cocoa production.Targeted training and knowledge transfer: Develop programs to support older farmers in adopting best agronomic practices through tailored technical training, facilitated input access, and advisory services, while also encouraging generational renewal among younger farmers.Sustainable intensification: Increase productivity on existing plots through cocoa-based agroforestry, while discouraging unsustainable expansion into remaining forest areas. Enhanced access to inputs and services: Improve the availability of quality planting material, fertilizers, credit, and agricultural advisory services, particularly for smallholders and older farmers who face barriers to adoption. Enhanced access to inputs and services: Improve the availability of quality planting material, fertilizers, credit, and agricultural advisory services, particularly for smallholders and older farmers who face barriers to adoption. Inclusive land policies and gender equity: Promote policies recognizing and supporting women’s contributions to cocoa production. Inclusive land policies and gender equity: Promote policies recognizing and supporting women’s contributions to cocoa production. Targeted training and knowledge transfer: Develop programs to support older farmers in adopting best agronomic practices through tailored technical training, facilitated input access, and advisory services, while also encouraging generational renewal among younger farmers. Targeted training and knowledge transfer: Develop programs to support older farmers in adopting best agronomic practices through tailored technical training, facilitated input access, and advisory services, while also encouraging generational renewal among younger farmers. Sustainable intensification: Increase productivity on existing plots through cocoa-based agroforestry, while discouraging unsustainable expansion into remaining forest areas. Sustainable intensification: Increase productivity on existing plots through cocoa-based agroforestry, while discouraging unsustainable expansion into remaining forest areas. Strengthening the resilience and sustainability of cocoa farming systems in Côte d’Ivoire requires coordinated actions addressing technical, institutional, and ecological constraints. By integrating these dimensions and focusing on the needs of older farmers, stakeholders can enhance productivity, reduce vulnerability, and support the long-term sustainability of the cocoa sector. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1