Integrating global crop models with knowledge graphs: Evolution, comparison, and future directions

Crop models leverage mathematical methods to simulate plant growth and environmental responses, forming a powerful and critical tool for agricultural decision support in research, management, and policy–making. However, a comprehensive review that systematically integrates global crop model resources with knowledge graphs linking their evolution to applications remains absent. Here, following a review of mainstream crop models, we constructed a multidimensional comparison framework to offer a more comprehensive overview, covering application areas, structural features, functional modules, and applicable crop types. We systematically integrated mainstream international crop models databases by classified key crop models into three groups based on their primary drivers and system scales. Mechanistic crop models focus on internal photosynthetic physiology. Process–based crop models simulate the soil–plant–atmosphere continuum, emphasizing resource movement and water–use efficiency. Integrated system crop models operate at the ecosystem level, addressing long–term dynamics like crop rotation, soil biogeochemistry, and management impacts. Subsequently, we performed a bibliometric analysis of relevant literature from 2013 to 2024 using CiteSpace. This analysis identified three primary research hotspots for crop models, i.e. enhancing yield simulation capabilities, assessing climate change impacts, and developing decision support systems for production management. Finally, we identified key limitations and prospective trends. The fundamental constraints are: (1) the disruption of model–reliant steady–state climate parameters by extreme weather; (2) the challenges of cultivar parameterization and local calibration; and (3) insufficient mechanistic understanding of pest/disease/weed dynamics, along with uncertainties in coupling model modules. Despite these challenges, we identify three development directions that are each supported by both the model comparison and the bibliometric analysis: remote-sensing data assimilation, integration with machine learning, and cross-cutting directions covering digital twins and multi-model coupling.