Evolution of crop models: Modeling approaches and software development

Plant-physiological crop yield models, referred to here as 'crop models', have undergone significant evolution over the past century, and studying this evolution is crucial for identifying the limitations of prior approaches and understanding how advancements overcome them. This study presents this evolution along two major dimensions, elucidating their historical, current, and future trajectories. The first dimension is the modeling approach, which has evolved from simple regression models to complex process-based models, and more recently to machine-learning based models. The second dimension is the crop model software development, which has evolved in terms of software architecture and development practices. This study shows that, in the first dimension, the current trajectory indicates increased mechanistic detail among the process-based models. Similarly, in the second dimension, model development has adopted decentralized development and modular architecture with a focus on user-centric design. The future trend is towards a highly interoperable development environment, powered by AI, with a focus on accessibility.