在植被定量遥感领域，辐射传输模型是理解植被冠层与太阳辐射之间相互作用过程以及发展植被参数反演算法的重要基础。目前，植被参数的遥感反演主要依靠一维辐射传输模型所建立的遥感观测与植被参数之间的数学关系，但是一维模型往往存在过多假设而与真实三维场景差别太远。三维辐射传输模型是解决该问题的有效手段，然而由于地表高度异质性，要发展一个能够描述三维真实场景且效率高的辐射传输模型，前提是要解决计算量大以及精细三维植被结构难以获取的问题。目前大部分三维模型都对场景进行了一定程度的简化或者只模拟小范围真实场景。虽然近年来在计算机图形学领域出现了一系列的基于物理原理的渲染引擎，但是他们并不是专门为科学计算而设计，很难直接应用于遥感信号模拟。 尽管如此，计算机图形学相关技术的发展以及计算机计算能力的提升为大尺度复杂场景下三维辐射传输模型的发展提供了更多的可能。此外，随着遥感数据获取能力的提升，高分辨数据（如无人机影像）以及三维数据（如激光雷达）越来越丰富。如何充分利用这些信息并与辐射传输模型相结合来获取地表特征参数将是未来遥感建模中的重要一环，而三维辐射传输模型正是这样的理想工具。三维辐射传输模型可以精确刻画任意复杂场景的辐射信号，能够更大限度地利用多源遥感数据所提供的信息量，对参数化物理模型的发展和验证具有重要作用。 本文的研究内容主要集中在两部分： （1）三维真实场景辐射传输模型构建 针对三维辐射传输模型的构建，本文建立了以光线追踪为基础的大尺度辐射传输模型LESS（LargE-Scale remote sensing data and image Simulation framework）。在该模型中，充分利用了光线前向追踪和后向追踪模式的特点，实现了从可见光到热红外的多波段、多角度、多分辨率遥感信号的模拟。其中，前向光子追踪算法用来模拟多波段多角度反射率因子（Bidirectional Reflectance Factor，BRF）及能量平衡相关参量（如上下行辐射）的计算。本文提出了虚拟光子算法，可利用较少的光子数量得到准确的BRF值。后向路径跟踪算法用来模拟各种投影模式下的相机成像过程。后向追踪算法显著降低了模拟过程中的内存消耗，可快速模拟大尺度遥感影像。为方便使用，LESS提供了一套易用的GUI（Graphical User Interface）界面以及相关的参数输入工具集。为了检验LESS模型模拟结果的精度，本文将LESS与目前主流的辐射传输模型以及实测数据进行了对比，结果显示，LESS兼具高的模拟精度和模拟效率。 （2）植被冠层三维结构信息提取及真实森林场景重建 针对三维辐射传输模拟中的场景生成问题，本文提出了基于单木和基于体元的两种森林场景重建方法。单木法以单木结构数据库和单木分割为基础，能够得到具有叶片尺度精细结构的三维场景。体元法以三维叶面积体密度为基础，将场景划分为规则的网格，并以网格为基本单元重建三维场景。冠层三维信息是三维辐射传输模型的基本输入，也是准确模拟三维复杂场景遥感信号的重要前提，其中最重要的参数就是叶面积指数（Leaf Area Index，LAI）和叶倾角分布（Leaf Angle Distribution，LAD）。针对这两种场景构建方法所需的植被结构参数，本研究提出了相应的参数获取算法。首先，本文重点研究了基于机载激光雷达回波强度信息的三维叶面积体密度反演算法，该算法仅利用地面回波点而无需对植被冠层有过多假设，从而具有较强的适应性。其次，本文也研究了基于智能手机照片的叶倾角分布提取方法，该方法能够利用低成本的智能手机照片以及手机自动记录的姿态信息准确快速获取叶倾角分布。 综上，本文从植被冠层三维信息获取和三维辐射传输模拟两个方面对冠层辐射传输建模过程进行了研究。所建立的LESS模型，可以为物理模型验证、参数化模型建立、神经网络和查找表建立等等提供可靠模拟数据集，促进新的模型建立和算法发展。

Understanding the detailed radiative transfer process between vegetation canopies and solar radiation is important for developing parameter retrieval algorithms in quantitative remote sensing of vegetation. Currently, one-dimensional (1D) radiative transfer models are most commonly used to obtain vegetation parameters by establishing a mathematical relationship between them and remotely sensed signals. However, 1D models usually have too many assumptions, which are far different from real canopy structures. To solve this problem, three-dimensional (3D) radiative transfer models are commonly used, however, developing an efficient 3D radiative transfer model, which can represent accurate and detailed canopy structures, is difficult, due to the complexity of the landscapes as well as the intensive computational cost of 3D radiative transfer simulations. To improve the efficiency, present models usually work with schematic landscapes or with small-scale realistic scenes. The computer graphics community provides the most accurate and efficient models (known as renderers), but they were not designed specifically for performing scientific radiative transfer simulations, which makes it difficult to be applied to remote sensing directly. Nevertheless, the development of rendering techniques and the improvement of computing power provide more possibilities for developing a 3D radiative transfer model which can be used to simulate large-scale scenes. Besides, as the development of data acquisition techniques, high-resolution datasets, i.e., images from unmanned aerial vehicle (UAV), and 3D information, i.e., Light detection and Ranging (LiDAR) data, become more and more accessible. How to use these datasets and combine them with 3D radiative transfer models for retrieving vegetation parameters will become one of the major topics in future study of remote sensing. Indeed, 3D radiative transfer models are ideal tools for this kind of task, because 3D models can represent arbitrary complex 3D scenes and can maximumly utilize the observations, which mitigates the “ill-posed” problem in remote sensing. Based on above analysis, this thesis focuses on: (1) Developing a 3D radiative transfer model based on realistic scenes This thesis proposes a ray-tracing based radiative transfer model, which is named as LESS (LargE-Scale remote sensing data and image Simulation framework). In LESS, we have taken full advantages of the forward and backward modes of ray-tracing technique to simulate remote sensing signals from visible to thermal infrared band. A virtual photon algorithm is proposed in forward mode to simulate BRF (Bidirectional Reflectance Factor) more efficiently with fewer photons, while an on-the-fly computation algorithm for determining the sunlit and shaded scene components is developed in backward mode to simulate thermal infrared images. LESS also provides a user-friendly graphic user interface (GUI) and a set of tools to help construct the landscape and set parameters. The accuracy of LESS is evaluated with other models as well as field measurements in terms of directional BRFs and pixel-wise simulated images, which shows very good agreement with high computational efficiency. (2) Retrieving 3D structural information of canopy for constructing realistic forest scenes. This thesis proposed two approaches to reconstruct realistic forest scenes: single tree based and voxel-based methods. Single tree approach is based on the establishment of a single tree database and the segmentation of single tree from images or LiDAR data. Voxel based approach divides the space into regular grids and retrieves leaf volume density (LVD) for each voxel from LiDAR data. 3D information of canopy is the basic input parameter for 3D radiative transfer model. Among all of these parameters, leaf area index (LAI) and leaf angle distribution (LAD) are the most import ones. For retrieving LAI, this thesis proposed an intensity-based method to retrieve 3D LVD from airborne LiDAR data. This method only considers the ground returns from each LiDAR pulse, regardless of the canopy structures. For LAD, this thesis proposed a method to measure two-dimensional (2D) LAD by using smartphone photos, which uses structure from motion (SfM) technique to recover 3D point cloud and the orientation information that is automatically recorded by smartphone to establish the relationship between local coordinates and world coordinates. In summary, this thesis focuses on the two aspects of the canopy radiative transfer modeling: 3D scene construction and 3D radiative transfer model development. LESS has the potential in simulating datasets of realistically reconstructed landscapes. Such simulated datasets can be used as benchmarks for validating physical models, developing parameterized models and training neural networks. Of course, 3D radiative transfer is a complex process, LESS still has the room for improvement in the future.