扩展三维辐射传输模型优化算法反演高分辨率叶面积指数和叶绿素含量

随着米级到十米级光学卫星的发展，能精确模拟冠层反射率的三维辐射传输模型（3D-RTM）在叶面积指数（LAI）和叶片叶绿素含量（LCC）反演中获得越来越多的关注。但受限于计算效率，3D-RTM很难快速构建包含大量不同土壤反射率、叶片生化参数组合的模拟数据集，用于机器学习或查找表等反演算法；而对输入参数的过度化简会影响模型反演精度和适用范围。为此，本项目以冠层反射率与波长、土壤反射率及叶片生化参数的物理关系为基础，扩展原本仅针对小麦和玉米的3D-RTM优化算法到作物与森林等不同植被类型，实现模拟数据集的快速构建。同时，结合深度学习算法和Sentinel-2反射率数据反演LAI和LCC。项目预期可得到能够实现不同植被冠层反射率模拟数据集快速构建的3D-RTM优化算法，以及基于3D-RTM优化算法的高空间分辨率LAI和LCC反演算法，对3D-RTM在参数反演、植被监测等方面的应用有重要意义。

With the increasing availability of frequent decametric and metric resolution satellite images, 3D radiative transfer model (3D-RTM) which can provide an accurate simulation of canopy reflectance is getting more and more attention in studies about inversion of leaf area index (LAI) and leaf chlorophyll content (LCC). However, it is time-consuming to use 3D-RTM to generate simulation dataset with large amounts of different combinations of soil reflectance and leaf biochemical variables for retrieval algorithms like machine learning and look-up-table, while the simplification of input parameters will reduce the accuracy of estimation and will limit the scope of the application. Therefore, according to the dependency of canopy reflectance on wavelength, leaf biochemical composition and soil reflectance, this project proposes to extend the specific 3D-RTM optimization algorithm of wheat and maize to different vegetation types including crops and forests, and to improve the computing efficiency of the generation of simulation database. Coupled with the deep learning method, the 3D-RTM optimization algorithm is used to retrieve LAI and LCC with Sentinel-2 observations. The expected outcome of this project including the 3D-RTM optimization algorithm which can speed up the generation of 3D radiative transfer simulation datasets of crops and forests and the estimated high spatial resolution LAI and LCC based on the 3D-RTM optimization algorithm. It also has important significance for the retrieval of vegetation variables and the monitoring of plants.