Estimating SPAD of tea leaf using UAV multispectral remote sensing based on stacking ensemble learning

Chlorophyll content (SPAD) is a key indicator of tea plant photosynthetic capacity and health. The remote sensing monitoring of tea chlorophyll has problems such as poor model stability and spectral saturation in areas with high chlorophyll values. Addressing the challenges of labour-intensive, inefficient, and destructive monitoring in hilly tea gardens, this study utilizes UAV multispectral remote sensing technology with field data from 2024 and 2025 to establish a rapid, non-destructive, and precise SPAD estimation model. The Otsu threshold of NDVI (0.14–0.52) effectively extracted tea planting areas. Correlation analysis revealed significant relationships between SPAD and multispectral indices such as NDVI, NDRE, and SIPI. Based on this, Partial Least Squares Regression (PLSR), Random Forest Regression (RFR), we propose a novel hybrid framework that systematically integrates PLSR’s linear feature extraction with RFR’s nonlinear residual learning via a shallow MLP meta-model (PLSR-RFR-Stacking model), specifically designed to overcome the spectral saturation limitation in high-chlorophyll tea canopies. Results showed that the PLSR-RFR-Stacking model, built on the combined two-year dataset, performed best. For modelling and validation, R2 values were 0.81 and 0.69, RMSE values were 4.43 and 5.70, and nRMSE values were 8.05% and 10.09%, respectively, demonstrating strong predictive power. Spatial distribution maps generated by the optimal model indicated that median SPAD values across six research areas ranged from 40.68 to 64.97, providing a quantitative reference for evaluating tea tree growth. This study proposes and validates a more robust and generalized SPAD inversion modelling framework, which solves the problem of poor cross‑year stability of the model, and promotes the application of UAV multispectral sensing in tea chlorophyll detection and supports the construction and precision management of digital tea gardens.