Leaf nitrogen content estimation in olive orchards based on analysis and modelling of UAV-acquired hyperspectral data

Nitrogen (N) management is critical for optimising yield and quality in modern super high-density (SHD) olive orchards while minimising environmental impacts. Traditional leaf nitrogen content (LNC) assessment relies on costly and time-consuming laboratory methods, lacking the resolution for precision agriculture. This study evaluated the potential of unmanned aerial vehicle (UAV)-based hyperspectral sensing to non-invasively estimate LNC in SHD olive groves. Hyperspectral images (537 bands, 400-2,500 nm) were acquired over an experimental orchard containing three olive varieties receiving contrasting N-fertigation treatments. Canopy spectral reflectance was extracted and subjected to preprocessing, including Savitzky-Golay filtering (first and second derivatives), scatter correction (MSC and SNV), and Principal Component Analysis (PCA). Partial Least Squares Regression (PLSR) and Artificial Neural Network (ANN) models were then trained to estimate LNC values derived from chemical analysis. Results from an external validation set showed that the ANN model provided superior performance compared to PLSR. The best-performing ANN model, which utilised normalised information from both spectral derivatives, achieved outstanding predictive accuracy ( R 2 > 0.8 and R P D = 2.3 ). This work demonstrates that a non-linear modelling approach leveraging UAV-acquired VNIR-SWIR hyperspectral data is a robust methodology for N status monitoring, offering the high spatial and temporal resolution required for precision fertilisation in modern olive cultivation.