Green Area Index (GAI) refers to the total one-sided area of green elements per unit ground surface area of the canopy and FIPAR is defined as the fraction of intercepted radiation by the canopy elements. These two variables characterize the interactions between the canopy and atmosphere and are key variables in many applications related to the vegetation functioning. The retrieval of GAI and FIPAR from remote sensing data can be performed either by using empirically (mainly based on vegetation indices) or physically based methods (radiative transfer model inversion). Regardless the type of method (Optimization, Look-Up-Table, Machine Learning), physically based algorithms generally require a significant number of simulations. Therefore, one-dimensional radiative transfer models have been mainly used since they are computationally efficient and characterized by a low number of inputs, which eases the setting of numerical experiments and constrains the possible ambiguities between variables during the inversion process. However, these models are based on the assumption that the canopy is a horizontally homogeneous and infinite turbid medium with leaves randomly dispersed. With the current numerical possibilities and considering the recent evolution of machine learning techniques, it is now possible to use more realistic and accurate radiative transfer models. The objective of this study is to quantitatively assess the gain in accuracy when using 3D model inversion as compared to 1D model for GAI and FIPAR estimation. We present a 3D radiative model combined with 3D architectural plant model to accurately simulate the canopy reflectance. This 3D model is run to build a simulated database that is later used by machine learning algorithms to estimate both GAI and FIPAR from reflectance data. We used LuxRender, a physically based ray tracing model and unbiased rendering engine to simulate reflectances from 3D mock-ups. We first evaluated LuxRender performances against the RAMI On-line Model Checker (ROMC). Then, we set up a database by running LuxRender on 3D mock-ups of wheat canopies at various phenologic stages and densities generated thanks to the Adel-Wheat model. The leaf optical properties were simulated with the PROSPECT model. In order to compare the impact of canopy structure assumption, both the proposed 3D radiative transfer model with realistic canopy architecture and SAIL model based on the turbid medium assumption are used to estimate GAI and FIPAR. Conclusions will be drawn according to the estimated results from these two methods.