Can Machine Learning Forecast Rice Yields in Data-Constrained Settings? Satellite Climate Data, National Crop Statistics, and Lessons from Sierra Leone
A new study published on arXiv explores the potential of machine learning to forecast rice yields in data-constrained environments, specifically focusing on Sierra Leone. Researchers found that models trained solely on crop statistics performed no better than simple persistence. However, when augmented with free satellite climate data, machine learning models, particularly XGBoost, significantly reduced forecast error by one-third, highlighting early-season rainfall as a key predictor. The study also noted that institutional factors, not climate, were responsible for a past yield collapse, and offers policy recommendations based on these findings. AI
IMPACT Demonstrates the potential of ML and climate data to improve agricultural forecasting in regions with limited data, offering actionable insights for policy.