Deep learning-based fusion of spectral and meteorological harmonic indices for monitoring tea plant carbon and nitrogen status under variable illumination conditions

Carbon (C) and nitrogen (N) accumulation are key indicators of tea plant growth status and yield potential. However, the instability of spectral responses under variable field illumination conditions and the insufficient consideration of long-term meteorological effects limit the accuracy and robustness of current monitoring approaches; therefore, this study aimed to integrate RapidSCAN CS-45 active optical sensing with meteorological information through a deep learning framework to achieve accurate monitoring of tea plant C and N status. Field experiments were conducted across multiple sampling sites in Jiangsu Province to systematically collect spectral data at stable (SIC) and non-stable illumination conditions (NIC), respectively. Meteorological data were concurrently obtained and processed using a harmonic decomposition technique to construct optimized meteorological harmonic index (OMHI). Tea plant biomass as well as C and N accumulation indicators were simultaneously determined for subsequent analysis and model development. Deep learning algorithms were used to combine spectral indices and OMHI to construct prediction models for tea biomass and C and N accumulation. The results demonstrated that the correlations between C and N indicators and spectral indices remained significant across varying illumination conditions. Specifically, the hybrid convolutional neural network (CNN) model incorporating spectral indices achieved R 2 values ranging from 0.70 to 0.83 and 0.67 to 0.84 under SIC and NIC, respectively. Further analysis revealed that OMHI provided superior explanatory power for leaf C and N parameters compared to traditional mean indicators, with r improving to 0.72–0.80 in spring and 0.82–0.85 in summer. Integrating OMHI with spectral indices using hybrid CNN significantly enhanced prediction accuracy, increasing R 2 values to 0.80–0.89 (SIC) and 0.68–0.87 (NIC). Furthermore, the deployment of the UAV platform expanded monitoring coverage, enabling precise characterization of spatial heterogeneity in tea plant growth. This study offers a robust technical framework and theoretical foundation for nutrient diagnosis and precision management in smart tea cultivation.