Cocoa is an unforgiving crop. The trees live for decades, and a successful harvest depends on an intricate, months-long sequence of flowering, pollination, and pod development. Each stage has precise water requirements, and a misstep at any point can ripple through the entire growing season. Too little rain at the wrong moment stresses the trees; too much rain, arriving while the trees are flowering or as young pods begin to form, can flood root systems and—critically—create ideal conditions for fungal diseases that thrive in wet conditions. Anna Lea Albright, who led the research as a postdoctoral fellow at the Harvard University Center for the Environment, explains that while drought’s dangers are well established, the other extreme has been dangerously underappreciated. Heavy rain during flowering, or rain that encourages the spread of fungal infection during flowering and early pod development, turns out to matter just as much.

The quantitative case is striking. In Ghana, the world’s second-largest cocoa producer, the researchers found that excess rain during the wet season, combined with dry-season drought, explains roughly two-thirds of the year-to-year swings in the national cocoa harvest. That is a remarkable degree of explanatory power for a crop whose fortunes are typically attributed to a bewildering mix of agronomic, economic, and political factors. The study, published in the Proceedings of the National Academy of Sciences, relied on two decades of newly available district-level data from the Ghana Cocoa Board, the government agency that oversees the industry. Because production grew steadily over that period as more land came into cultivation, the team stripped out each district’s long-term trend to isolate the year-to-year fluctuations they wanted to understand.

With those swings in hand, the researchers subjected them to a battery of statistical tests against daily weather records. Average rainfall, heavy downpours, dry spells, soil moisture, and temperature extremes were all in the lineup of possible culprits. The verdict was clear: the best predictors of a bad cocoa year in Ghana were heavy rain during the main wet season and too little rain during the dry season. In other words, it is not simply how much water falls on a cocoa farm in a year, but when it falls and how violently it arrives. Short, intense bursts of rain—exactly the kind of events that average rainfall statistics smooth over—turn out to inflict disproportionate damage.

The timing of these destructive rains is no accident of geography. In Ghana, the most damaging downpours arrive during the April-to-June wet season, precisely when cocoa trees are flowering and young pods are beginning to form—the crop’s most vulnerable window. Later, during the November-to-February dry period, insufficient rain can hurt the crop as pods continue to develop. Because cocoa is especially vulnerable at certain stages of its growth cycle, adaptation cannot be a blunt instrument; it must be precise. Disease control efforts, for instance, can be timed around periods of heavy rain, but only if farmers receive forecasts with enough lead time to act—a capacity the research team is now working to develop.

To test whether Ghana was an anomaly, the researchers widened their lens to two other major producers on different continents: Ecuador and Indonesia. The pattern held. Years with heavier wet-season rains tended to be worse for cocoa in both countries, suggesting that the vulnerability is not a local quirk of West African agriculture but a fundamental feature of how the crop responds to water extremes across the tropics. This consistency across continents strengthens the case that rainfall extremes deserve a central place in climate-risk assessments for global cocoa, alongside the heat and drought concerns that have dominated the literature to date.

The predictability angle is where the research becomes genuinely actionable. The team emphasizes that the rains that hurt cocoa are tied to larger climate patterns, such as El Niño and shifts in Atlantic sea-surface temperatures. Because these drivers operate on seasonal timescales, some bad years may be predictable months before the damage is done—early enough to give farmers a meaningful warning. The stakes are illustrated by current conditions: a strong El Niño is developing, an event that typically favors heavy rainfall in coastal Ecuador and drier weather across West Africa and Indonesia. A functioning early-warning system could, in principle, tell farmers in each region which of these risks is elevated for the coming season.

The consequences of getting this wrong are already visible on grocery shelves. The volatility documented in the study echoes through global markets, where raw cocoa prices tripled in 2024, a spike attributed in part to climate-driven supply shocks. Chocolate manufacturers have responded with higher prices, smaller bars, and reformulated products, meaning the costs of unpredictable rainfall are being borne not only by farmers but by consumers worldwide. Senior author Peter Huybers, chair of Harvard’s Department of Earth and Planetary Sciences, hopes the findings will help reduce future crop damage, possibly by encouraging more canopy cover to shield cocoa flowers from large raindrops, or by retaining more leaf litter to prevent the splashes that transmit infection from soil to cocoa pods. These are inexpensive, locally implementable measures—but they work best when farmers know wet conditions are coming.

The study comes with important caveats that the researchers are careful to acknowledge. Rain is not the only problem Ghanaian cocoa farmers face. Aging trees, gold mining encroaching on farmland, smuggling, fertilizer costs, pests, and disease can all depress production, and none of these factors can be addressed by a weather forecast. There is also a technical limitation: rainfall data do not always capture the biggest downpours well, and those are precisely the events that matter most for crop damage. And looking further ahead, the picture grows murkier still. Climate change has the potential to substantially alter the timing and pattern of rainfall in cocoa-growing regions, yet current climate models struggle to effectively represent even the present-day climatology of rainfall over West Africa, let alone project how it will change in coming decades.

Even so, the study performs a valuable service by clarifying the nature of the climate threat to chocolate. Cocoa’s future is not simply a story about a hotter world. It is a story about when the rain falls, how hard it falls, and whether the people who grow the crop can anticipate those extremes with enough warning to adapt. For decades, climate science has taught farmers to fear the sun and the drought. This research shows that the rain itself—arriving too hard, too fast, at the wrong moment—may be the more immediate and more tractable danger. Turning seasonal climate patterns into practical warnings for millions of smallholder farmers is now the challenge, and one that could determine whether the world’s chocolate supply grows more secure or more precarious in the years ahead.

Subject of Research: How rainfall extremes drive cocoa yield losses across the tropics and could be predicted in advance

Article Title: Chocolate’s new climate threat: Too much rain

Article References: Chocolate’s new climate threat: Too much rain. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: cocoa, chocolate, climate change, heavy rainfall, drought, West Africa, Ghana, El Niño, crop yields, fungal disease, seasonal forecasting, Harvard University

The research, published in Discover Agriculture, was led by John Atsu Agbolosoo of IPB University in Indonesia together with Collins Yeboah Asiedu of Ghana Cocoa Board, Dick Chune Midamba and Mary Kwesiga of Gulu University, Stephen Opoku-Mensah of Kumasi Technical University, and Benjamin Sarfo of Akenten Appiah Menka University. The team surveyed 241 smallholder cocoa farmers in the Dodi Papase area of Ghana’s Kadjebi District during the 2023/2024 crop year, drawing participants from Cocoa Health and Extension Division registers of farmers who had already carried out at least one rehabilitation activity. Their central insight is deceptively simple but methodologically consequential: farmers do not treat the four main rehabilitation techniques as mutually exclusive menu items. Instead, they mix, match, and sequence coppicing, phased replanting, planting seedlings under old trees, and complete replanting in ways that reflect their resources, constraints, and risk tolerances.

Previous studies, most of them conducted in Nigeria, typically analyzed rehabilitation decisions one technique at a time, using single-equation logit or probit models that treat each choice as independent. The authors of the new work argue that this approach can produce biased or inefficient estimates because the latent factors driving one decision, such as managerial skill, risk appetite, or informal information networks, often spill over into others. To capture that interdependence, they estimated a multivariate probit model, a framework that models four binary outcomes simultaneously and allows the unobserved error terms to be correlated through a full variance-covariance matrix. A likelihood ratio test of joint independence decisively rejected the null hypothesis, with a chi-squared statistic of 28.33 on six degrees of freedom and a p-value below 0.01, confirming that the four decisions are statistically linked and that the joint model is warranted.

The correlation structure itself proved revealing. The strongest positive association appeared between coppicing and complete replanting, with an estimated correlation coefficient of 0.589, followed by phased replanting and complete replanting at 0.510, both significant at the one percent level. Phased replanting and planting under old trees showed a moderate positive correlation of 0.330. By contrast, the link between planting under old trees and complete replanting, at 0.171, was not statistically significant, hinting that these two strategies may respond to different unobserved drivers. The authors caution that these correlations reflect associations among unobserved determinants rather than causal complementarities, but they nonetheless paint a picture of farmers weighing strategies within a shared decision framework shaped by labor, liquidity, and information.

Adoption patterns were strikingly uneven. Coppicing, the practice of cutting mature trees back to stumps to stimulate fresh shoots while preserving shade cover and soil stability, was the most popular choice, practiced by 58 percent of surveyed farmers. Planting seedlings under aging trees, an interplanting approach that allows households to keep earning from old stands while hybrids mature, followed at 53 percent. Phased replanting, which renews plots section by section to smooth income losses, was used by 44 percent, while complete replanting, the most capital and labor intensive option suitable for severely degraded farms, trailed at 37 percent. Most tellingly, only about 12 percent of farmers adopted all four practices, and just 2 percent adopted none, indicating that selective combinations rather than wholesale adoption define rehabilitation behavior in the region.

The determinants of these choices were far from uniform. Farmer age was positively associated with coppicing, phased replanting, and under-planting, but showed no significant link with complete replanting, suggesting older farmers gravitate toward gradual, lower-cost strategies. Education displayed the opposite pattern in an unexpected direction: it was positively associated with complete replanting but negatively associated with coppicing and under-planting, implying that schooling shapes which technique is chosen rather than simply whether rehabilitation happens. Gender mattered too, with female farmers significantly less likely to adopt phased replanting and under-planting, a pattern the authors interpret as consistent with constraints on land access, labor arrangements, and input distribution that affect women disproportionately. Household composition also proved nuanced. Active household labor, rather than raw household size, was the variable that counted, being positively linked with under-planting and negatively with coppicing.

Institutional and incentive variables told an equally differentiated story. Cooperative membership was strongly and positively associated with coppicing, while access to planting materials from the Cocoa Health and Extension Division was significantly linked only with complete replanting. Record keeping and information access pushed farmers toward complete replanting but away from coppicing, underscoring that information does not uniformly raise adoption but instead steers farmers among alternatives. Training, surprisingly, was negatively associated with phased replanting, under-planting, and complete replanting, a counterintuitive result the authors attribute cautiously to program design or targeting effects that cannot be untangled with cross-sectional data. Farm size was positively associated with phased replanting and under-planting but negatively associated with coppicing and complete replanting, and off-farm income activity correlated positively with three of the four strategies, likely by financing rehabilitation investments while drawing labor away from the most demanding option.

The choice of study region gives these findings added urgency. The Oti Region and neighboring Volta Region have watched official cocoa output collapse, with production dropping by more than 6,000 tonnes over four years and roughly 7,128 tonnes lost to smuggling into Togo between 2020 and 2025. Poor roads to remote farms make official buying centers hard to reach, pushing farmers toward quick illicit payments, and the resulting low recorded volumes starve the area of the fertilizer supplies, extension support, and youth programs that COCOBOD allocates based on tonnage. Unlike Western North, Western South, and Ashanti, which attract the bulk of research attention and intervention funding, Oti remains fertile and suitable for cocoa but largely neglected. The authors argue that understanding rehabilitation behavior in such marginal areas is essential if Ghana’s national cocoa strategy is to extend beyond its traditional heartlands.

The methodological rigor of the study strengthens its conclusions. The multivariate probit specification was chosen over multinomial alternatives because it handles correlated, non-mutually-exclusive binary outcomes while remaining computationally tractable for mid-sized agricultural surveys. Diagnostic checks confirmed the model’s suitability: mean variance inflation factors of 1.83 ruled out multicollinearity, robust standard errors corrected heteroscedasticity detected in the complete replanting and under-planting equations, and normality tests were satisfied for three of the four outcome equations. The Wald chi-squared statistic of 582.14 on 72 degrees of freedom confirmed that the full set of eighteen explanatory variables was jointly significant.

For policymakers at Ghana Cocoa Board and its Cocoa Health and Extension Division, the message is that one-size-fits-all rehabilitation programs are unlikely to match how farmers actually behave. The evidence points toward strategy-specific support: subsidized planting materials for those contemplating full replanting, cooperative-based channels for coppicing, targeted labor and land-access interventions for women, and credit or input schemes that ease the liquidity crunch of capital-intensive options. The authors are careful to note that with cross-sectional data and a sample restricted to farmers who have already rehabilitated, the findings describe associations rather than causal effects, and they call for panel studies, inclusion of non-adopters, and qualitative work to unpack the cultural and gender dynamics beneath the statistics. Still, in a sector where every hectare of diseased, abandoned farmland represents lost income for families and lost foreign exchange for a nation, knowing that farmers choose rehabilitation paths strategically rather than uniformly is a genuinely useful piece of the puzzle.

Subject of Research: Statistical modeling of cocoa farm rehabilitation strategy adoption among smallholder farmers in the Oti Region of Ghana

Article Title: Using a multivariate probit method to model decisions on cocoa farm rehabilitation in Ghana

Article References: Agbolosoo, J. A., Asiedu, C. Y., Midamba, D. C., Opoku-Mensah, S., Sarfo, B., & Kwesiga, M. (2026). Using a multivariate probit method to model decisions on cocoa farm rehabilitation in Ghana. Discover Agriculture, 4(1), Article 283. https://doi.org/10.1007/s44279-026-00765-3

Image Credits: AI Generated

DOI: 10.1007/s44279-026-00765-3

Keywords: cocoa, Ghana, farm rehabilitation, multivariate probit, smallholder farmers, coppicing, replanting, CSSVD, agricultural economics, technology adoption, COCOBOD, Oti Region