Above-ground biomass (AGB) is a key indicator of crop growth and yield formation, and its accurate estimation is essential for maize growth monitoring and field management. In this study, unmanned aerial vehicle (UAV) hyperspectral reflectance data collected at three maize growth stages were used to estimate the AGB parameter. The Grünwald-Letnikov fractional-order differentiation (FOD) method was then applied to generate candidate spectral vegetation index feature sets with fractional orders ranging from 0 to 2 at an interval of 0.2. Based on a unified training/testing partition, three PLSR-based strategies were constructed, including a fixed strategy ( v = 0), a globally optimal fractional-order strategy, and a stage-specific optimal fractional-order strategy. For each strategy, the top ten features ranked by variable importance in projection (VIP) were selected for model construction, and model performance was evaluated using R², RMSE, and nRMSE. The results showed that fractional-order optimization improved AGB estimation performance compared with the fixed strategy ( v = 0), indicating that appropriate fractional-order transformation can enhance the expression of AGB-sensitive spectral information. Among the three strategies, the stage-specific optimal fractional-order strategy achieved the best overall performance (R² = 0.91 and RMSE = 131.14 (g/m 2 )), and showed strongest adaptability across different growth stages, especially during the middle and late stages. Moreover, by further comparing the predicted and observed AGB distribution results, the stage-specific optimal fractional-order strategy exhibited the best distribution-matching ability across all three growth stages, indicating that this strategy could more effectively preserve and distinguish the differences among plots and samples. These findings demonstrate that stage-specific fractional-order optimization is more effective than unified fractional-order strategies for multi-stage maize AGB estimation and has strong potential for refined biomass monitoring and crop growth assessment.
Wang et al. (Mon,) studied this question.