Journal of Agronomy and Field Innovations

Background: Maize (Zea mays L.) is a cereal crop of strategic importance for global food, feed and bioenergy security, but its productivity is highly susceptible to climatic variability, soil heterogeneity and recurrent foliar and stalk diseases. Traditional machine learning (ML) methods for yield forecasting and disease prediction have demonstrated high predictive accuracy, but remain largely “black box” methods, which has limited their uptake by agronomists, extension officers and smallholder farmers. 
Objective: In this work, we propose and evaluate an Explainable Artificial Intelligence (XAI) framework that integrates multi-source agronomic data with interpretable machine learning and deep learning architectures to predict maize yield and disease risk, while providing transparent and actionable explanations. 
Methods: Data were collected from multi environment field trials with weather data, soil physicochemical properties, vegetation indices derived from unmanned aerial vehicle (UAV) and satellite data, Internet of Things (IoT) sensor data streams and ground truth disease and yield observations. Ensemble tree-based models, convolutional neural networks, and hybrid models were applied for benchmarking. Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were used to interpret feature contributions. 
Results: The gradient-boosted ensemble with SHAP interpretation achieved the best yield prediction accuracy (R 2 = 0.93; RMSE = 0.41 t ha −1) and disease forecasting performance (F1-score = 0.91), outperforming traditional deep learning baselines and still being interpretable. The most important predictors across environments were canopy temperature, normalized difference vegetation index (NDVI), relative humidity and nitrogen status. 
Significance: The XAI framework significantly increased the model transparency without compromising the predictive performance, producing decision-support recommendations that can be directly linked to the underlying agro-environmental drivers. 
Conclusion: Explainable Artificial Intelligence offers a scientifically sound and operationally reliable route towards climate-smart, precision maize production, bridging the gap between algorithmic sophistication and field-level decision making.