A novel UAV-based fusion of spectral and textural features with canopy stratification for accurate estimation of vertical nitrogen distribution in cotton

Accurate monitoring of plant nitrogen accumulation (PNA) is crucial for precision management in cotton, a crop with a complex, multi-layered canopy. Existing unmanned aerial vehicle (UAV) remote sensing approaches, predominantly reliant on top-of-canopy spectral vegetation indices (VIs) like NDVI, suffer from signal saturation and fail to capture the vertical heterogeneity of nitrogen within the canopy. To better characterize these limitations, this study developed and rigorously validated an innovative framework that synergistically integrates multi-source UAV features with a vertical canopy stratification strategy. We conducted a comprehensive water-nitrogen coupling field experiment and extracted key spectral indices (e.g., NDRE, GNDVI) and textural features (e.g., Entropy, Contrast). Three machine learning models were evaluated under stratified and whole-plant scenarios. The results demonstrate that the fusion of VIs and textural features within a stratified framework leads to distinct estimation performance across different canopy layers. The Random Forest Regression model applied to the upper canopy layer achieved the highest performance (R 2 = 0.833, RMSE = 1.463 g m⁻ 2 ), compared with an R 2 of 0.494 obtained from the conventional whole-plant approach. Furthermore, the stratification strategy enabled a quantitative characterization of dynamic nitrogen distribution patterns within the cotton canopy, with relatively uniform allocation at the budding stage and pronounced concentration in the photosynthetically active middle and upper layers during the boll-setting stage (e.g., middle-layer PNA reaching 121.84 kg ha⁻ 1 ). This study provides a three-dimensional perspective on canopy nitrogen distribution that helps link UAV-based observations with physiologically meaningful processes and offers a potential pathway for informing precision nitrogen management in cotton and other crops with complex canopy architectures.