Model pruning enables localized and efficient federated learning for yield forecasting and data sharing.