Integrated diagnostics and time series sensitivity assessment for growth monitoring of a medicinal plant (Glycyrrhiza uralensis Fisch.) based on unmanned aerial vehicle multispectral sensors

Ao Zhang¹Haibin Guan¹Heng Zhi Dong²Jia Xin¹Yan Xue¹Yu Feng Han³Jiang Ling Meng³Ling Xiu Yu³Qin Xiao Wang^1*Cao Yang^1*

• ¹College of Pharmacy, Inner Mongolia Medical University, Hohhot, Inner Mongolia Autonomous Region, China
• ²Department of Pharmacy, Inner Mongolia People's Hospital, Hohhot, Inner Mongolia Autonomous Region, China
• ³Inner Mongolia Shuangqi Pharmaceutical Co Ltd, Hohhot, Inner Mongolia, China

Background: Water and nitrogen are essential elements prone to deficiency during plant growth. Current water-fertilizer monitoring technologies are unable to meet the demands of large-scale Glycyrrhiza uralensis cultivation. Near-ground remote sensing technology based on unmanned aerial vehicle (UAV) multispectral image is widely used for crop growth monitoring and agricultural management and has proven to be effective for assessing water and nitrogen status. However, integrated models tailored for medicinal plants remain underexplored. Methods: This study collected UAV multispectral images of G. uralensis under various water and nitrogen treatments and extracted vegetation indices (VIs). Field phenotypic indicators (PIs), including plant height (PH), tiller number (TN), soil plant analysis development values (SPAD), and nitrogen content (NC), were synchronously measured. Models were constructed using backpropagation neural network (BP), support vector machine (SVM), and random forest (RF) to evaluate PIs to predict yield and monitor growth dynamics. Yield predictions based on PIs were further compared with validate model performance. Results: The results demonstrated that both the RF algorithm and excess green index (EXG) exhibited versatility in growth monitoring and yield prediction. PIs collectively achieved high-precision predictions (mean 0.42 ≤ R 2 ≤ 0.94), with the prediction of PH using green leaf index (GLI) in BP algorithm attaining peak accuracy (R² = 0.94). VIs and PIs exhibited comparable predictive capacity for yield, with multi-indicators integrated modeling significantly enhancing performance: VIs achieved R² = 0.87 under RF algorithms, whereas PIs reached R² = 0.81 using BP algorithms. Further analysis revealed that PH served as the central predictor, achieving R² = 0.74 under standalone predictions of RF algorithm, whereas other parameters primarily enhanced model accuracy through complementarity effects, thereby providing supplementary diagnostic value. Conclusions: This research established a high-precision, time-efficient, and practical UAV remote sensing-based method for growth monitoring and yield prediction in G. uralensis, offering a novel solution for standardized production of medicinal plant resources.