Answers are grounded in this paper and cite their sources.
Accurate and timely crop yield prediction is essential for ensuring global food security, optimising agricultural resource allocation and informing policy decisions in an era of accelerating climate change. The rapid growth of satellite remote-sensing archives, high-resolution weather and soil datasets, and advances in machine learning (ML) and deep learning (DL) have led to new approaches with promise at regional and national scales. This review examines the current state of ML and DL applications in crop yield prediction, covering classical ML; deep learning, including Transformers; transfer learning; hybrid and ensemble models; explainable AI; and uncertainty quantification. Classical ML approaches include Random Forest, XGBoost and Support Vector Machines, while deep learning architectures include CNNs, LSTMs and CNN-LSTM hybrids. Emerging Transformer-based models use self-attention to model long-range temporal dependencies, while transfer learning strategies support adaptation across regions and crop varieties. Explainable AI (XAI) techniques, including SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), attention-based attribution and counterfactual explanations, help bridge the gap between predictive accuracy and agronomic interpretability. Uncertainty quantification methods include Bayesian neural networks, Monte Carlo dropout, quantile regression and conformal prediction. Several shortcomings remain, including the lack of standardised benchmark datasets, inconsistent evaluation practices that may inflate reported performance, and limited provision of calibrated prediction intervals. The unresolved question is not simply which algorithm achieves the highest accuracy on a given dataset, but whether these models can reliably predict yields under challenging climatic conditions. This review outlines directions for benchmarking, uncertainty-aware modelling and the integration of explainability into decision-support tools for agricultural crop yield prediction.
No takes yet. Share an insight, caveat, or question.
Vamsi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: