星载光子计数激光雷达森林叶面积指数反演方法研究

The accurate estimation of forest leaf area index is very important for understanding the material and energy exchange process of terrestrial ecosystem. Previous studies have demonstrated the possibility of space-borne LiDAR in retrieving forest leaf area index. Given space-borne photon-counting LiDAR is quite different from the Geoscience Laser Altimeter System (GLAS) in many respects, such as detection mechanism, data record type and spatial distribution pattern, there is an urgent need to design new methods of data processing and leaf area index retrieval for space-borne photon-counting LiDAR data. Therefore, this research aims to propose effective algorithms for processing space-borne photon-counting LiDAR data, and explore the possibility of space-borne photon-counting LiDAR data in estimating forest leaf area index. Firstly, we will analyze the generation mechanism and spatial distribution of noise photons, and propose an effective noise removal algorithm based on multilevel analysis for filtering out noise photons. Secondly, a new photon classification method based on moving curve fitting will also be proposed to effectively divide the de-noised photons into canopy photons, ground photons and noise photons in the condition of steep terrain and dense vegetation. Finally, a series of photon characteristic parameters relating to the canopy structural information will be designed and extracted from space-borne photon-counting LiDAR data, and the inversion models for the estimation of leaf area index will be established based on the empirical model and canopy gap probability model, respectively. Additionally, we will also quantitatively evaluate the influence of positioning error, data signal-to-noise ratio, forest type, terrain, canopy closure and other factors on the inversion models. This study will provide not only effective data processing algorithms for the new generation satellite LiDAR system, but also new models and methods for the accurate estimation forest leaf area index.

精确估算森林叶面积指数对理解陆地生态系统物质和能量交换过程有着重要的意义。已有研究表明星载激光雷达GLAS能够精确提取森林叶面积指数，但鉴于星载光子计数激光雷达在探测机理、数据记录方式与空间分布规律等方面与GLAS存在差异，适用于星载光子计数激光雷达的数据处理与叶面积指数反演方法有待探索。本项目拟深入分析光子噪声产生机理与空间分布规律，研究并提出基于多层级渐进的光子点云去噪算法；构建基于移动曲线拟合的光子点云分类方法，实现在复杂地形及密集覆盖下植被光子、地面光子和噪声光子的有效分离；结合星载光子计数激光雷达数据的大光斑、点云、剖面分布等特点，构建表征冠层结构信息的特征参数并建立森林叶面积指数反演模型，评估光子点云定位误差、数据信噪比、森林类型、地形、郁闭度等因素对反演模型精度的影响。项目成果将为新一代星载激光雷达数据处理提供算法基础，也为森林叶面积指数的高精度反演提供新的模型与方法支撑。

激光雷达（Light Detection And Ranging, LiDAR）具有很强的植被冠层穿透能力，能够直接获取高精度的植被三维结构信息。其中星载LiDAR由于其覆盖范围广、数据获取成本低等优势，已成为叶面积指数（LAI）精确提取的有效手段。虽然已有研究表明星载全波形LiDAR能够精确提取森林LAI，但鉴于星载光子计数LiDAR在探测机理、数据记录方式与空间分布规律等方面与星载全波形LiDAR存在差异，已有森林LAI反演模型与方法并不适用于星载光子计数LiDAR。因此，本项目旨在研究星载光子计数LiDAR数据处理技术和森林LAI反演模型，包括以下三个主要研究内容：1）光子点云去噪算法研究；2）光子点云分类方法研究；3）基于星载光子计数激光雷达ICESat-2数据的森林LAI反演研究。.通过上述研究取得了以下重要结果和数据产品：1）提出了基于改进OPTICS的光子点云去噪算法，实现了强背景噪声下的信号光子的精确提取，相比已有算法精度更高；2）提出了基于卷积神经网络的ICESat-2光子点云分类方法，提高了复杂地形和密集植被覆盖下的光子点云分类精度；3）开展了不同模态星载LiDAR森林高度一致性分析研究，构建了森林高度一致性模型，为融合不同模态星载LiDAR进行森林参数制图提供了可靠支撑；4）构建了基于星载光子计数激光雷达ICESat-2数据的森林LAI反演模型；5）获取了ICESat-2数据、无人机LiDAR点云数据、去噪与分类后的光子计数LiDAR数据、野外实测森林参数数据、森林LAI产品等科学数据与产品。.本项目取得的成果为星载光子计数LiDAR数据处理提供了新的算法，也为森林LAI的高精度反演提供了可靠的模型与方法。