The current research focuses on studying rice production, which accounts for about 62% of the national energy
intake, as an important element of food security in Sierra Leone. Currently, there is no objective approach to predicting rice
yields in a timely manner. In this work, we propose a machine learning model that uses multi-temporal Sentinel-2 images,
the ERA5-Land and CHIRPS weather datasets, and climate forecasting datasets to predict early-season rice yields for
Kambia and Bombali districts in Sierra Leone. A training dataset with 1,774 samples of bi-weekly data collected during
seven rice-growing seasons (2018-2024) is used to optimize an ensemble consisting of Ridge Regression, Random Forest, and
XGBoost.