Advancing soybean biomass estimation through multi-source UAV data fusion and machine learning algorithms

Abstract

Technological advances in unmanned aerial vehicle (UAV) systems offer significant potential for the rapid and efficient monitoring of soybean aboveground biomass (AGB) in precision agriculture, providing an alternative to traditional AGB measurement techniques. However, recent studies have indicated that relying solely on vegetation indices (VIs) can lead to inaccurate AGB estimations due to variability in crop cultivars, growth stages, and environmental conditions. This study evaluated the performance of UAV-derived features (including canopy spectral, textural, and structural features) in estimating AGB across fifty soybean cultivars and multiple growth stages in a two-year field experiment, utilizing various machine learning algorithms (decision tree, DT; random forest, RF; neural network, NN; extreme gradient boosting, XGBoost; and ensemble learning, EL). The findings revealed that: (1) The integration of UAV digital imagery with the canopy height model (CHM) facilitated the estimation of soybean plant height, with the coefficient of determination (R²) and root mean square error (RMSE) values for ground-measured and UAV-derived plant height across different growth stages ranging from 0.72 to 0.88 and 3.35 to 6.13 cm, respectively. (2) Textural and structural features demonstrated good sensitivity to AGB variability across cultivars and growth stages, despite each feature type having its limitations. The fusion of UAV-derived spectral, textural, and structural features yielded the highest accuracy (R² = 0.85), significantly improving model performance compared to using dual (R² ranging from 0.79 to 0.81) feature types. (3) Model accuracy significantly varied across different growth stages. For machine learning algorithms, the EL model outperformed DT, RF, NN, and XGBoost in AGB prediction, consistently providing accurate estimations multiple soybean growth stages. These findings highlight the potential of integrating multi-source UAV features to enhance soybean AGB estimation, facilitating farmers decision-making in precision crop management and assisting breeders to select high- and sustainable-yielding cultivars in large-scale breeding program.