Monitoring rice grain protein accumulation dynamics based on UAV multispectral data

Context or problem: The rapid and accurate estimation of grain protein accumulation (GPA) dynamics in rice by remote sensing (RS) is essential for rice grain quality assessment. However, challenges are faced when multi -spectral data are directly used to estimate the GPA due to the lack of corresponding sensitive feature bands. Objective or research question: The purpose of this study is to establish an estimation model that can monitor the dynamics of grain protein accumulation after rice heading. It aims to solve the monitoring of dynamic indicators (such as GPA) in the process of grain quality formation. Methods: To achieve progress in GPA monitoring, physiological parameters that are closely connected to GPA must be considered as an intermediate bridge between RS and GPA. The indirect model of "spectral parameter -physiological parameter -protein accumulation dynamic" was used in this study to develop a RS method for monitoring the dynamics of GPA based on a machine learning method. The GPA three-parameter estimation model was established using three physiological parameters as intermediate parameters: leaf nitrogen content (LNC), leaf nitrogen accumulation (LNA), and nitrogen harvest index (NHI) with the leaf N status. Results: The results show that the indirect model outperforms the direct estimation model of protein accumu-lation based on the vegetation index in terms of prediction accuracy. Furthermore, the GPA three-parameter estimation model was tested using dataset from different multispectral cameras (Mini-MCA6 and AIRPHEN), revealing the good performance with R2 and RMSE of 0.71 and 14.86 g / m2 respectively. Conclusions: These results indicated that incorporating nitrogen status and nitrogen partitioning into the GPA RS estimation model could improve the estimation accuracy of grain protein accumulation dynamics in rice. Implications or significance: The success in estimating GPA with this method would advance the timely monitoring of different crop productivity indicators during the formation process. This study has great potential for improving RS monitoring of grain quality formation processes and facilitating spectral estimation of grain quality.