Optimizing nitrogen estimates in common bean canopies throughout key growth stages via spectral and textural data from unmanned aerial vehicle multispectral imagery

Abstract

Leaf nitrogen assessment is crucial for optimizing crop management, driving remote sensing use. This study demonstrates the effectiveness of unmanned aerial vehicles (UAVs) multispectral imagery for enhancing leaf nitrogen content estimation in common bean (Phaseolus vulgaris L.) through the integration of vegetation indices (VIs) and texture features. Research conducted over two years (2021–2022) evaluated various nitrogen rates across critical growth stages (V4, R5, and R7). Machine learning models combining spectral and textural information significantly outperformed single-index approaches, achieving root mean square error (RMSE) values of 1.80 g kg−1 (relative root mean square error – RRMSE = 2.93 %) at V4 stage using support vector machine with VIs, and 2.79 g kg−1 (RRMSE = 5.20 %) at R5 stage using random forest with VIs. For later growth stages (R7) and across the entire season (all growth stages), the combination of VIs and texture metrics proved most effective, with random forest achieving RMSE values of 3.42 and 3.96 g kg−1 (RRMSE = 7.40 and 7.32 %), respectively. Texture analysis in across-row directions (90° and 135°) provided superior performance compared to traditional diagonal approaches for row-planted crops. Linear regression analysis showed that normalized difference texture indices incorporating correlation and homogeneity explained up to 71 % of leaf nitrogen content variability at R7 stage. The optimal nitrogen rate of 91 kg ha−1, validated through both yield response and leaf nitrogen measurements, provides a robust benchmark for nitrogen management in common bean production. This methodology offers a practical framework for real-time, site-specific nitrogen management that improves upon current recommendation systems.