A Federated Learning Framework for Sustainable Boro Rice Yield Prediction Using Multi-Regional Agro-Climatic Data

Boro rice, the primary dry-season crop in Bangladesh, plays a critical role in national food security. Accurate prediction of yield is critical in the context of escalating climate uncertainty and mounting population pressure. This study will propose a federated machine learning model that leverages 25 years (2000-2024) of agro-climatic and production data across six major divisions of Bangladesh to predict Boro rice yield. A novel dataset was built from satellite-extracted environmental characteristics and ground-truth yield information from the Bangladesh Bureau of Statistics and NASA POWER. The local-level performance of ten regression models was compared, in which SGD regressor was the best performer(up to R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.978). A Federated Averaging (FedAvg) approach was used for decentralized learning with linear…