Field Crop 2025, Vol.8, No.6, 293-300 http://cropscipublisher.com/index.php/fc 295 2.3 Advantages: non-invasive, large-scale monitoring, real-time data acquisition Compared with sampling and measuring each plant one by one, the benefits of remote sensing are too obvious. It can obtain large-scale field data without destructive sampling or a large amount of manual labor (Omia et al., 2023). Moreover, one shot is not enough; it can be taken repeatedly, which is both fast and extensive, allowing you to almost instantly observe the changes in the health of the crops. In this way, both disease signs and nutrient deficiencies can be detected in advance (Li et al., 2023). When this type of data is combined with models or machine learning, the monitoring accuracy and scalability will be further enhanced. For agricultural managers, this means they can intervene earlier, use land more rationally, and achieve the most stable harvest with the least input. 3 Crop Modeling Techniques for Wheat Yield Prediction 3.1 Overview of process-based models: APSIM, DSSAT, WOFOST In wheat yield prediction, several frequently mentioned models are almost household names: APSIM, DSSAT, and WOFOST (Wajid et al., 2021). Their common point is that they all attempt to use daily meteorological, soil and crop data to restore the entire process of plants from emergence to panicle formation, including growth, development and biomass accumulation. The difference lies in that the algorithmic details of each model are slightly different, and thus their applicable scopes are not exactly the same. Some studies suggest that the prediction results of DSSAT and DAISY are more stable, while APSIM and WOFOST also perform well in complex environments. When it comes to which one is the best, there is actually no definite conclusion. So some scholars simply combine multiple models to make up for the bias of a single model. This kind of "combination punch" is often more stable when dealing with uncertainties. 3.2 Role of input variables: weather, soil data, crop genotype Whether the model is accurate or not mainly depends on the data "fed" in. Meteorological data, soil properties and variety characteristics - all three are indispensable. The meteorological section generally includes temperature, precipitation, radiation intensity, etc. In terms of soil, it involves texture, nutrient and moisture characteristics; Crop genotypes reflect variety differences. Studies have found that the availability of water and nitrogen is most likely to influence the yield outcome, and precipitation, soil type and nitrogen fertilizer application rate are often sensitive factors (Hao et al., 2024). Of course, if fine calibration can be made for the soil and varieties in different regions, the error will be significantly reduced, especially in arid or barren environments, which is more prominent (Shahid et al., 2024). 3.3 Output utility: predicting phenology, biomass, and grain yield The output results of these models are actually very useful. Not only do they provide the final grain yield, but they can also predict the flowering period, maturity period, leaf area index (LAI), and even changes in aboveground and underground biomass. For researchers, this helps to understand the response of wheat under different climatic and management conditions; For farmers or policymakers, these data can be used to plan resources and adjust planting strategies. Furthermore, if the model is combined with remote sensing images or machine learning, real-time yield predictions with geolocation can be generated (Kheir et al., 2023), which makes the application of precision agriculture more operational and also provides a basis for food security planning. 4 Synergistic Integration of Remote Sensing and Crop Models 4.1 Data assimilation methods to feed real-time remote data into models Data assimilation is a crucial step for the model to "understand" the real-time situation in the field. In simple terms, it is to continuously stuff remote sensing observations into the model to keep the model's state in sync with reality. The commonly used methods include ensemble Kalman filter (EnKF), four-dimensional variational assimilation (4DVar), particle filter, as well as particle swarm optimization (PSO) or Complex Evolutionary Algorithm (SCE-UA), etc. (Jin et al., 2022). These algorithms can dynamically correct the variables in the model based on the leaf area index (LAI) or soil moisture data sent back by satellites. For instance, when MODIS or Sentinel data is input into WOFOST or SAFY models, the predicted yields are often much more accurate than before, with a significant decrease in error, whether at the field or regional scale.
RkJQdWJsaXNoZXIy MjQ4ODYzNA==