Rice Canopy Nitrogen Content Estimation Using UAV and XGBoost

An operational framework for the estimation of rice Canopy Nitrogen Content (CNC) at KALRO Kirogo farm in Kenya's Mwea Irrigation Scheme is proposed in this paper. The proposed framework integrated UAV multispectral imagery with XGBoost. A DJI Phantom 2 Multispectral drone captured imagery at 50 m altitude, achieving 5 cm spatial resolution. Five Vegetation Indices (Vis) (NDRE, ReCI, GCI, NRI, GLI) were extracted from radiometrically corrected orthomosaics and paired with Soil and Plant Analysis Development (SPAD) measurements from stratified random sampling across three growth stages (Vegetative Stage (VS), Early Maturity (EM), and Late Maturity (LM)). XGBoost hyperparameters were optimized via grid search (144 combinations, 5-fold CV), achieving superior performance. SHAP analysis identified NDRE as the primary predictor with agronomically sensible contribution. High-resolution nitrogen maps enabled the detection of within-field variability at management-relevant scales for smallholder plots. The results demonstrated that UAV-XGBoost integration provides accurate, non-destructive nitrogen estimation for sub-Saharan African smallholder rice systems, offering a scalable framework for precision nitrogen management.