Next-generation hybrid precipitation forecasts that integrate Indigenous knowledge

Abstract Many smallholder farmers in the Global South utilize local forecasts based on Indigenous knowledge due to limited reliability and accessibility of scientific weather forecasts. The use of local forecast, however, faces challenges by increasing climate variability, which undermines farmers’ confidence in their forecast. This study addresses these challenges by developing a hybrid forecast that integrates both scientific and local forecast using machine learning techniques to improve precipitation predictions in northern Ghana. Results show that the hybrid forecast improves precipitation forecast accuracy by 23% and 33% compared to scientific forecast and local forecast, respectively. The best performance is achieved by combining two random forests (RFs) or a voting classifier and a RF. This research highlights the potential of machine learning to develop more accurate hybrid forecast than other statistical methods. Such enhanced precipitation forecasts could enable smallholder farmers in the Global South to make better-informed agricultural decisions, ultimately enhancing their livelihoods.