From uncertainty to resilience: A systematic review and meta-analysis of data-driven climate adaptation in African cocoa production

Background Climate change threatens African cocoa production. Côte d'Ivoire faces 27–50% area loss by mid-century. Ghana faces 17–23% output decline. Digital agriculture tools have tripled across Africa. Evidence on their resilience impacts remains sparse. Objectives This systematic review and meta-analysis examined associations between data-driven climate adaptation interventions and cocoa resilience outcomes in Africa. Secondary objectives included testing conditional heterogeneity and assessing evidence gaps. Methods A PRISMA-compliant search was conducted across Web of Science, Scopus, CAB Abstracts, EconLit, Google Scholar, IEEE Xplore, ACM Digital Library, grey literature, and French databases (inception to April 2026). PICOS criteria specified cocoa smallholders, data-driven interventions, yield or profit or stability outcomes, and quasi-experimental or panel designs. Random-effects meta-analysis was performed on adjudicated yield effects. The a priori moderator hypotheses were not tested because the moderator dataset could not be reconciled with the adjudicated effect matrix. Risk of bias was assessed using ROBINS-I. GRADE certainty was evaluated. Results From 2847 records, 23 studies met inclusion criteria (21 quasi-experimental, 2 panel, 0 RCTs). The primary quantitative synthesis was restricted to 12 adjudicated yield effects from 12 independent studies. The pooled standardised effect was 0.454 (95% CI 0.382 to 0.526), with no detectable heterogeneity (τ 2 = 0, I 2 = 0%). Yield-stability and profit effects were retained descriptively because they were heterogeneous and not commensurate. The pooled estimate is a standardised association among a restricted set of comparable estimates; it is not a causal effect and should not be interpreted as a percentage yield increase. Formal moderator and meta-regression analyses were not reported because the moderator dataset could not be reconciled with the adjudicated effect matrix; the a priori moderator hypotheses therefore remain untested. Limitations The primary synthesis contained only 12 independent studies, which limits precision and power for moderator and publication-bias assessment. The evidence base was geographically concentrated (Ghana k = 13, Côte d'Ivoire k = 4, Nigeria k = 4, Cameroon k = 2). No RCTs were identified. GRADE certainty was low to very low. Formal small-study-effect testing was not considered sufficiently reliable given the limited number of independent studies and the heterogeneity of estimands and designs. Conclusion Data-driven adaptation showed a positive standardised association with yield in observational studies from Ghana, Côte d'Ivoire, Nigeria, and Cameroon. This association should not be interpreted as an unconditional causal effect. Causal inference was not possible due to the absence of RCTs and serious risk of bias in included studies. Policy recommendations were limited to pilot programs with embedded rigorous evaluation. Major evidence gaps included environmental outcomes, profit measurement, and data from under-represented producer countries.