Cocoa; future yields across West Africa

世界可可产量的70%由西非的小农户生产。然而，产量很低，面临气候变化、病虫害的风险，许多可可种植家庭生活在国际贫困线附近或以下。需要进行重大变革，以发展可持续的供应链。能够准确地预测可可产量和生产将是这样一个变化。目前，只有少数几家大型跨国可可公司和经纪人能够获得准确的产量预测，这些公司和经纪人根据对数百个可可农场的实际监测运行自己的专有预测方案。可可业的所有其他行为者都缺乏关于预期收成的可靠信息来源。然而，这种预测对于支持整个供应链的规划和适应至关重要。它们对价格形成也至关重要，而价格形成是由期货市场决定的，期货市场依赖于参与者对未来收成的预期。目前，市场上的信息不对称，有利于少数大玩家。在这里，我们建议使用一种经过验证的技术，以专有系统成本的一小部分生成准确的预测模型，使可可行业的广大受众能够负担得起，从而为可可行业的民主化做出贡献。我们将联合收割机现有的一系列长期（20-40年）可用的数据源与机器学习和模型相结合，以生成强大的预测工具。我们现在有了产量关键决定因素的时间序列图，包括降雨量、叶面积指数、表层土壤湿度、土壤物理性质。我们也有区域和国家的产量数据。有强有力的证据表明气候（土壤湿度）对产量的影响，但关于干旱胁迫影响的临界阈值的信息很少。在这里，我们使用过程建模将植物生产与生根区的土壤水分联系起来。我们校准我们的模型在现场规模耦合植物-土壤过程，然后校准模型在区域和国家尺度耦合产量植物-土壤指标。其结果是一个强大的工具，用于生成全年的产量预测，并具有明确的置信区间。这些信息为可可价值链提供了新的输入。正如两个项目合作伙伴所证明的那样，这项研究的输出将构成商业提供商向可可行业最终用户提供产品和服务的科学基础。一旦在可可上得到证明，这种方法就可以在一系列其他作物上复制，从谷物到热带商品再到植物油。

Seventy percent of the world cocoa production is produced by small holders in West Africa. However, yields are low and at risk from climate change, pests and diseases and many cocoa farming families live around or below the international poverty line. Major changes are needed to develop a sustainable supply chain. Being able to accurately predict cocoa yields and production will be one such change. Currently, accurate production forecasts are only available to a handful of large multinational cocoa companies and brokers, who run their proprietary forecasting programmes, based on physical monitoring of hundreds of cocoa farms. All other actors in the cocoa industry lack a reliable source of information about the expected harvest. However, such forecasts are essential to support planning and adaptation across the supply chain. They are also essential to price formation, which is determined in futures markets, which rely on participants' expectations of future harvests. Currently, information is asymmetrical in the market, favouring the few big players. Here we propose to use a demonstrated technology to generate accurate forecasting models at a fraction of the cost of the proprietary systems, making them affordable for a broad audience in the cocoa industry and hence contributing to the democratisation of the cocoa sector. We combine a range of data sources now available over long periods (20-40 years) with machine learning and models to generate a robust forecast tool. We now have time series mapping of critical determinants of yield, including rainfall, leaf area index, surface soil moisture, soil physical properties. We also have regional and national yield data. There is strong evidence of climate (soil moisture) impacts on yield, but little information on the critical thresholds in drought stress effects. Here we use process modelling to link plant production to soil moisture through the rooting zone. We calibrate our model at site scale to couple plant-soil processes, and then calibrate the model at regional and national scale to couple yield to plant-soil indicators. The outcome is a robust tool for generating forecasts of yield through the year, with clear confidence intervals. This information provides the novel input for the cocoa value chain.The output from this research will form the scientific basis on which commercial providers can offer products and services to end-users in the cocoa industry, as demonstrated by two project partners. Once proven for cocoa, the approach can be replicated in a range of other crops, from grains to tropical commodities to vegetable oils.