A Spatio-Temporal Transformer Framework for High-Resolution Crop Yield Prediction Across Agro-Climatic Zones: Survey

Accurate crop yield prediction is a cornerstone of modern precision agriculture, enabling proactive decision-making for food security, supply-chain optimization, and climate-resilient farming. Traditional statistical models and conventional machine learning approaches are limited in their ability to model complex, nonlinear, and long-range dependencies inherent in agro-climatic systems. Recent advances in deep learning, particularly transformer-based architecture, have demonstrated remarkable performance in modeling spatio-temporal dependencies across diverse domains. This paper presents an extensive study on crop yield prediction using a spatio-temporal transformer framework capable of integrating multi-modal datasets such as satellite imagery, weather observations, soil characteristics, and historical yield records. A comprehensive literature survey is conducted to examine existing machine learning, deep learning, and attention-based models. Based on the identified research gaps, multiple transformer-driven methodologies are proposed with mathematical formulations. Comparative analysis using reported metrics from prior studies highlights the superiority of transformer-based models in terms of accuracy, scalability, and robustness across agro-climatic zones.