Biomass prediction and shoot growth characterization of single-staked yam plants using UAV imagery

). Multi-angle aerial images from nadir and oblique views were used to extract vegetation- and height-related indices that served as predictors in machine learning models. Support vector regression using combined-view imagery provided the highest prediction accuracy (R² = 0.79) and remained robust across growth stages, years, fertilizer treatments, and genotypes. Notably, the combined-view configuration outperformed single-view imaging, demonstrating the advantage of capturing complementary canopy-structure information in complex staked-vine canopies. Time-series biomass estimates enabled the fitting of genotype-specific Richards growth curves using Bayesian inference. Significant genotypic variations were observed in parameters associated with maximum biomass and early growth rate, whereas phenology-related parameters showed comparatively minimal differences. These parameter differences may reflect variation in canopy architecture and growth allocation among genotypes. Overall, this integrated workflow provides a scalable tool for nondestructive monitoring of yam growth dynamics and for summarizing biomass trajectories with interpretable parameters, supporting breeding efforts aimed at improving yam productivity and yield stability across diverse cultivation conditions.