Remote sensing is needed for better managing vegetation covers. Hence, three-dimensional (3D) radiative transfer (RT) modeling is essential for understanding remote sensing signals of complex 3D vegetation covers. Due to the complexity of 3D models, one-dimensional (1D) RT models are commonly used to retrieve vegetation parameters, e.g., leaf area index (LAI), from remote sensing data. However, 1D models are not adapted to actual vegetation covers because they abstract them as schematic 1D layers, which is not realistic. Much effort is devoted to the conception of 3D RT models that can consider the 3D architecture of vegetation covers. However, developing an efficient 3D RT model that works on large and realistic scenes is still a challenging task. Major difficulties are the intensive computational costs of 3D RT simulation and the acquisition of detailed 3D canopy structures. Therefore, 3D RT models usually only work on abstracted scenes or small realistic scenes. Scene abstraction may cause uncertainties, and the small-scale approach is not compatible with most satellite observations (e.g., MODIS). The computer graphics community provides the most accurate and efficient models (i.e., renderers). However, the initial renderer models were not designed for accurate RT modeling, which explains the difficulty to use them for remote sensing applications. Recently emerged advanced techniques in computer graphics and light detection and ranging area (LiDAR) make it more possible to solve the above problems. 3D RT can be greatly accelerated due to the increasing computer power and improvement of rendering algorithms (e.g., ray-tracing acceleration and computational optimization). Also, 3D high-resolution information from LiDARs and photogrammetry become more accessible to reconstruct realistic 3D scenes. This approach requires new processing methods to combine 3D information and 3D RT models, which is of great importance for better remote sensing survey of vegetation. This thesis is focused on 1) Development of a 3D RT model based on recent ray-tracing techniques and 2) Retrieval of 3D leaf volume density (LVD) for constructing 3D forest scenes. This first chapter presents the development of an efficient 3D RT model, named LESS (LargE-Scale remote sensing data and image Simulation framework). LESS makes full use of ray-tracing algorithms. Specifically, it simulates multispectral BRF and scene radiative budget with a weighted forward photon tracing method, and sensor images (e.g., fisheye images) or large-scale (e.g. 1 km2) spectral images are simulated with a backward path tracing method. In the forward mode, a "virtual photon" algorithm is used to simulate accurate BRF with few photons. The backward mode is used to simulate thermal infrared images and also atmosphere RT. LESS efficiency and accuracy were demonstrated with a model intercomparison and field measurements. In addition, LESS has an easy-to-use graphic user interface (GUI) to input parameters, construct and visualize 3D scenes. 3D forest reconstruction is done with a simulated LiDAR dataset to assess approaches that retrieve LVD from airborne LiDAR data. The dataset is simulated with the discrete anisotropic radiative transfer model (DART).