Rapid modeling of 3D rice canopy structure considering vertical heterogeneity and analysis of spectral response

The vertical heterogeneity of rice canopy structure limits the accuracy of inverting leaf physicochemical parameters using traditional radiative transfer models, while LiDAR-based 3D reconstruction remains costly for large-scale applications. To address these challenges, this study proposes a method for constructing 3D rice canopy scenes using “Precision Mode” and “Rapid Mode” strategies. The Precision Mode builds detailed structural models based on measured morphological parameters, validated via the LESS 3D radiative transfer model. To overcome the limitations of obtaining detailed morphology via UAV remote sensing, the Rapid Mode employs machine learning algorithms—specifically Support Vector Machine (SVM), Random Forest (RF), and XGBoost—to map easily accessible parameters (LAI, Above-ground Biomass, Plant Height, and Transplanting Date) to detailed 3D structural parameters. Results indicate that XGBoost achieves the highest accuracy in the Rapid Mode. Furthermore, simulated spectra under both modes showed high consistency with measured spectra, yielding average RMSE values of 0.0104 (R² = 0.9965) for the Precision Mode and 0.0307 (R² = 0.9694) for the Rapid Mode. Although the spectral accuracy of the Rapid Mode is slightly lower, its modeling efficiency is significantly enhanced, retaining a strong capability to reproduce spectral response characteristics across growth stages. This approach provides an effective tool for analyzing vertical spectral response mechanisms and offers an efficient data simulation scheme for UAV remote sensing parameter inversion based on 3D radiative transfer models.