Abstract:

 Timely and accurate access to crop yield information is critical for decision−making in national food policy and safety assessments. Remote sensing technology, with low cost and high efficiency, provides an effective means for large−scale crop yield estimation. Strengthening the integration of agronomic knowledge into crop yield estimation models and addressing the scarcity of training samples are key challenges in current research. In this paper, authors synthesized existing literature, summarized data− and knowledge−driven methods for crop yield estimation, and systematically discussed the methods for constructing multi−scenario simulation datasets and modelling techniques for crop yield estimation based on hybrid modeling approaches. The paper also provided an overview of commonly used models and algorithms, and summarized the application of remote sensing technology to hybrid modeling methods. Finally, it comprehensively discussed the uncertainties in hybrid modeling and outlined the future trends and challenged in crop yield estimation studies. The results showed that hybrid modeling approaches, driven by both data and knowledge had made significant progress in crop yield estimation. By combining the advantages of data−driven and knowledge−driven models, these approaches reduced the reliance on ground−truth samples while enhancing the mechanistic support for the predictions. Limiting factors for improving the accuracy of crop yield estimation included uncertainties in remote sensing data sources, the uncertainties in knowledge−driven models when simulating crop physiological processes, and the uncertainties in the predictions of data−driven models. Future trends will focus on improving the quality and availability of input data, strengthening the theoretical foundation of knowledge−driven models, and advancing algorithm improvements in data−driven models.

Key words: Remote sensing, Machine learning, Crop growth model, Hybrid modeling