High-Resolution Maize Yield Mapping across Africa using Earth Observation and Machine Learning, Deep Learning, and Foundation Model

Abstract Africa’s food security is increasingly threatened by climate change and population growth. High-resolution yield data are vital for precision agriculture and climate adaptation, yet much of the continent lacks sufficient monitoring due to limited ground data.This study presents first high-resolution (250 m), continent-wide maize yield prediction framework for 42 African countries and a novel yield disaggregation method using Net Primary Productivity (NPP) to spatially downscale national-level FAO yield statistics, creating fine-scale training data for supervised learning. A comprehensive feature set of 296 variables was constructed by integrating multi-source Earth observation, climate, and soil data. The framework evaluates multiple machine learning and deep learning models- including XGBoost, LightGBM, a hybrid deep neural network (HDNN), and, for the first time in this context, the Tabular Prior-data Fitted Network (TabPFN), a tabular foundation model. Using an expanding-window temporal cross-validation strategy, XGBoost achieved the highest temporal R² (0.78), while TabPFN demonstrated superior spatial generalization and the lowest mean absolute percentage error (MAPE ≈ 25%). Causal inference and ablation analyses underscored the predictive importance of vegetation indices (e.g., NDVI, NDWI), drought metrics, and soil properties. Model outputs showed strong alignment with FAOSTAT-reported national yields (R² > 0.75; MAPE ≈ 26–28%), highlighting the reliability of the proposed approach. Despite known limitations- such as reliance on proxy-based disaggregation and the use of coarse-resolution climate inputs- this work provides a novel and scalable framework for yield monitoring in data-scarce regions. It also marks the first application of tabular foundation models in continental-scale agricultural prediction, opening new directions for high-resolution, data-efficient crop yield prediction.