Probabilistic yield forecasting of robusta coffee at the farm scale using agroclimatic and remote sensing derived indices

• Forecasting capability of coffee yield models at the farm scale was assessed. The modelling approach involved Bayesian and machine learning methods. Potential predictors included data, agroclimatic remotely sensed indices. Coffee yields were predicted with reasonable accuracies. In-season models’ forecast skills depend on data availability selected predictors. Timely reliable forecasts using information are pivotal to success agricultural climate risk management throughout value chain. statistical different lead times during growing season Using collected a 10-year period (2008-2017) from 558 farmers across four major coffee-producing provinces in Vietnam (Dak Lak, Dak Nong, Gia Lai, Lam Dong), built through robust involving Overall, estimated accuracies study based agroclimate variables, satellite-derived actual evapotranspiration, crop information. Median values prediction mean absolute percentage error (MAPE) ranged generally 8% 13%, median root square errors (RMSE) between 295 kg ha −1 429 . For one month before harvest, did not vary markedly when comparing MAPE RMSE values. farms Dong, forecasting varied 13% 16% 420 456 , respectively. readily freely available explored this appears flexible for an application larger number Vietnamese regions. Moreover, can serve as basis developing predicting system that will offer substantial benefits entire industry better supply chain countries worldwide.