This study integrates a dynamic plant growth model with a three-dimensional (3D) radiative transfer model (RTM) for maize traits retrieval using high spatial-spectral resolution airborne data. The research combines the Discrete Anisotropic Radiative Transfer (DART) model with the Dynamic L-System-based Architectural maize (DLAmaize) growth model to simulate field reflectance. Comparison with the 1D RTM SAIL revealed limitations in representing row structure effects, field slope, and complex light-canopy interactions. Novel Global Sensitivity Analyses (GSA) were carried out using dependence-based methods to overcome limitations traditional variance-based approaches, enabling better characterization of hyperspectral sensitivity to changes in leaf biochemistry, canopy architecture, and soil moisture. GSA provided complementary results to assess estimation uncertainties of the proposed traits retrieval method across growth stages. A hybrid inversion framework combining DART simulations with an active learning strategy using Kernel Ridge Regression was implemented for traits estimation. The approach was validated using ground data and HyPlant-DUAL airborne hyperspectral images from two field campaigns in 2018 and achieved high retrieval accuracy of key maize traits: leaf area index (LAI, R2=0.91, RMSE=0.42 m2/m2), leaf chlorophyll content (LCC, R2=0.61, RMSE=3.89 mu g/cm2), leaf nitrogen content (LNC, R2=0.86, RMSE=1.13 x 10-2 mg/cm2), leaf dry matter content (LMA, R2=0.84, RMSE=0.15 mg/cm2), and leaf water content (LWC, R2=0.78, RMSE=0.88 mg/cm2). The validated models were used to generate two-date 10 m resolution maps, showing good spatial consistency and traits dynamics. The findings demonstrate that integrating 3D RTMs with dynamic growth models is suited for maize trait mapping from hyperspectral data in varying growing conditions.

来源平台：REMOTE SENSING OF ENVIRONMENT