Field Crop 2025, Vol.8, No.6, 293-300 http://cropscipublisher.com/index.php/fc 298 greatly reduce the value of information (Wang et al., 2024). However, there are also many problems on the model side: once the input data such as meteorology, soil, and management are incomplete, both the inversion accuracy and the simulation results will be affected (Chen and Tao, 2022). This situation is particularly evident in areas where ground monitoring is weak and production environments vary greatly, and the uncertainty of data is often further magnified. 7.2 Technical complexity in integration, requiring interdisciplinary expertise Integrating remote sensing with crop models is no easy task. It requires the collaboration of knowledge from multiple fields such as agriculture, remote sensing, data science, and modeling. Data assimilation requires the use of ensemble Kalman filtering (EnKF), 4DVar or machine learning algorithms, and the subsequent calibration and verification processes cannot be taken lightly (Zare et al., 2024). To run models on a regional scale, one still has to deal with huge parameterization tasks and computational pressure. Without a stable data processing procedure and a professional team, it is difficult to do this job well. For some regions with insufficient technological reserves, this interdisciplinary requirement itself has become a bottleneck for promotion. 7.3 Limited access to high-resolution models or sensor data in resource-poor regions In developed regions, satellites, drones and hyperspectral sensors have long been common, but in areas with limited resources, these devices remain "luxuries". Not only are the purchase and maintenance costs high, but processing and storing big data also require expensive computing power and technical support (Dhakar et al., 2022; Joshi et al., 2023). In addition, due to the lack of high-resolution crop models with open access and the scarcity of datasets for machine learning training, many regions simply cannot run such integrated systems. The result is that the technology seems advanced, but there is still a long way to go before it can be truly popularized. 8 Future Directions and Concluding Remarks The emergence of artificial intelligence and machine learning is quietly rewriting the way agricultural output is predicted. Nowadays, satellite images, weather records, soil information, and even field management data can all be integrated into a self-adjusting model. Unlike traditional single models, those "hybrid" methods that combine process models such as APSIM with machine learning algorithms (like random forests, XGBoost, and deep neural networks) are more like letting data and mechanisms speak together. The results also prove that this approach is often more stable and accurate, especially showing advantages in environments with variable climates or significant management differences. They can capture complex nonlinear changes, identify trends in advance, and provide stronger support for early prediction. Next, the academic community is increasingly focusing on how to further integrate deep learning and data assimilation technologies, so that "mechanism-driven" and "data-driven" can truly complement each other. Another direction worth paying attention to is to bring these technologies out of the laboratory and make them more "user-friendly". Now there are open cloud platforms and Web decision-making systems based on Google Earth Engine. Users do not need powerful computing devices to obtain production prediction results online. They integrate remote sensing, meteorological, management and other data on one interface, enabling farmers, agricultural technicians and even government officials to quickly obtain useful information. Meanwhile, low-cost open-source remote sensing tools and simple software also give regions with limited resources the opportunity to use these technologies, helping small-scale farmers reduce risks and improve management efficiency. Of course, promoting these integrated technologies is not something that can be achieved overnight. To make it truly effective, continuous investment is still needed in data infrastructure, interdisciplinary collaboration and talent cultivation. Only in this way can the integration of AI, remote sensing and modeling be implemented in a broader agricultural system, thereby supporting climate-resilient production, enhancing early warning capabilities, and ultimately providing reliable support for global food security and sustainable development. Acknowledgments We would like to thank the anonymous reviewers and the editor for their suggestions on terminology consistency, which improved the text's presentation.
RkJQdWJsaXNoZXIy MjQ4ODYzNA==