森林是陆地生态系统的主体，其既是全球气候变化的响应主体之一；也是实现碳中和目标，进行全球气候干预调节的媒介之一。如何快速准确地获取冠层结构参数对监测陆地碳循环和全球变化等具有重要意义。在众多的冠层结构参数中，冠层高度和 LAI （Leaf Area Index）是林学、生态学和地理学等多领域常用的森林状态指示变量。与已有的ICESat-1/GLAS （Ice, Cloud, and land Elevation Satellite /Geoscience Laser Altimeter System） 星载全波形雷达系统相比，全新一代的星载光子计数 LiDAR 系统（ICESat-2/ATLAS, Advanced Topographic Laser Altimeter System）在光斑尺寸和轨道密度上的优势使其具备在全球尺度下进行更高空间分辨率的冠层结构参数反演能力。然而，这一能力却极大地依赖于 ATLAS数据的处理，即光子点云去噪和分类。已有的光子点云去噪和分类算法仍不成熟，这使得在此基础上的冠层结构参数反演仍存在较大的不确定性。本研究旨在针对 ICESat-2/ATLAS 光子点云数据提出一种鲁棒的光子点云处理算法，在此基础上实现冠层高度和 LAI的反演。本文主要从以下三个方面开展研究：（1）自适应光子点云去噪算法和高鲁棒性的光子点云分类算法研究；（2）基于 ATLAS 数据的冠层高度反演及其影响因素分析；（3） 结合多光谱数据的冠层LAI反演。

本文的主要研究内容和结论如下：

（1）提出了一种针对 ICESat-2 ATLAS 数据的多阶段参数自适应的光子点云去噪算法。该算法包含三个相互耦合并逐层递进的去噪阶段，各阶段针对性地去除特定类型的噪声光子。具体来说，基于邻域点云距离和的粗去噪阶段针对冠层上部和地面下部的噪声；基于改进 LOF （Local Outier Factor）模型的精去噪阶段聚焦于靠近冠层顶部和地面下部以及冠层内部噪声；基于 Tukey 箱须图的残余噪声光子去除阶段则用于去除上述两个阶段残留的极少数噪声光子。三个去噪阶段共同实现噪声光子去除。结果表明，本文去噪算法能够有效地应用于具有异质噪声密度的数据，且具备一定的冠层内部噪声光子检出能力。与已有算法相比，本文算法的自适应特性使其可以有效地应用于不同地形坡度、不同植被类型和不同获取时间（日间和夜间）场景下的光子点云去噪。

（2）提出了改进的二维渐近不规则三角网加密算法，用于进行 ICESat-2 ATLAS光子点云分类。基于地面光子具有较好坡度连续性的事实，研究利用潜在地面光子的坡度和密度信息剔除地面下部残余噪声光子，同时利用密度方向信息剔除密度冠层区域冠层下部植被点误分带来的伪地面点，实现了地面种子的准确提取。最后利用 PTD （Progressive Triangular Irregular Network Densification） 模型和经典的滑动窗口模型分别实现了地面光子和冠层顶部光子的提取。与 ATL08 算法相比，本文算法能够更为完整地提取地面光子，能够有效地避免地面下部残余噪声光子和高植被覆盖度区域中冠层底部光子误分导致的地面光子错误提取，为 ATLAS数据的应用提供坚实的数据基础。

（3）基于 ICESat-2 ATLAS数据实现了地面高程、冠层顶部高程和冠层高度反演，并探讨了地形坡度和冠层覆盖度对冠层高度反演的影响。结果表明，基于本文算法的地面高程（MAE = 0.53m）和冠层顶部高程（MAE = 3.02m）与 ALS参考值的一致性明显高于 ATL08 算法的地面（MAE = 1.19m） 和冠层顶部高程（MAE = 3.64m）。本文算法的冠层高度与基于机载离散激光雷达 （Airborne LiDARScanner，ALS） 参考值的一致性亦明显优于 ATL08 算法结果，尤其是在低信噪比的日间数据中，本文算法的精度提升显著，R’从 ATL08 的0.26提升为0.60。针对坡度和冠层覆盖度对冠层高度反演误差的分析表明，本文算法对地形坡度和冠层覆盖度的依赖明显小于 ATL08 算法；ICESat-2 数据更适用于中高植被覆盖度区域或时相（≥0.4）的冠层高度反演。不同生态站点的冠层高度反演结果表明，ICESat-2更适用于大冠幅树种的冠层高度反演。

（4）基于 ICESat-2 ATLAS 数据实现了多种生态站点的冠层 LAI 反演。反演过程充分考虑了异质噪声密度和冠层内部噪声光子对冠层雷达穿透指数的影响。同时结合 Landsat 8 OLI 多光谱数据实现了冠层雷达穿透指数到冠层间隙率的校正，最后使用间隙率模型实现了冠层 LAI 的反演。与相应的 ALS LAI 结果对比表明，ATLAS 能够有效地应用于不同生态站点的冠层 LAI反演（R=0.54 ～0.94），并在较大尺度下（～100m）获得最优结果。与 MODIS LAI 的对比结果表明，ATLAS能够显著改善被动光学遥感在密集冠层（LAI≥4.2）中的 LAI 低估问题。总体而言，ATLAS 具备生产全球尺度 LAI产品的能力，这将是有别于被动光学遥感 LAI数据的全新产品。

Forests are the main body of terrestrial ecosystems，which is one of the main bodiesresponding to global climate change and also one of the mediums to realize the goal ofcarbon neutrality and to regulate global climate interventions. How to quickly and ac-curately obtain canopy structure parameters is of great significance for terrestrial carboncycle and global change monitoring. Among many canopy structure parameters, canopyheight and leaf area index （LAl） are commonly used as forest status indicator variablesin several fields. Compared with the existing spaceborne full-waveform LiDAR systems，ICESat-1/GLAS （Ice, Cloud, and land Elevation Satellite / Geoscience Laser Altime-ter System）， the new generation of spaceborne photon-counting LiDAR systems （ICESat-2/ATLAS, Advanced Topographic Laser Altimeter System） has the advantage of spot sizeand orbit density, which makes it capable of inverting canopy structural parameters withhigher spatial resolution on a global scale. This capability, however, relies heavily onthe processing of ATLAS photon spot cloud data, i.e.， noise filtering and classification ofphotons.The aim of this study is to propose a robust photon point cloud processing algorithmfor ICESat-2/ATLAS photon point cloud data, on the basis of which the inversion ofcanopy height and LAl for a variety of ecological sites in the United States is realized. Thispaperfocuses on the following three aspects： （1） adaptive photon point cloud denoisingalgorithm and highly robust photon point cloud classification algorithm； （2） canopy heightinversion based on ATLAS data and its influencing factors analysis; and （3） canopy LAIinversion combined with multispectral Landsat & OLI data.

The main research content and conclusions of this paper areas follows：

（1） A multi-stage parameter-adaptive denoising algorithm for ICESat-2/ATLAS datawas proposed. The algorithm consists of three progressive denoising stages, each of which was targeted to remove specific types of noise. The coarse denoising stage based onthe sum of distance to neighboring points targets the noise upper the canopy and belowthe ground； the fine denoising stage based on the improved LOF （Local Outlier Factor）model focuses on the noise near to the top of the canopy and the ground， as well as thenoises within the canopy; and the residual noise photon removal stage based on the TukeyBoxplot was used to remove the residual noise photons surviving the above two stages.The results showed that the denoising algorithm in this paper can be effectively applied todata with heterogeneous noise densities and can detect noise photons within the canopyto some extent. Compared with the existing algorithms, the proposed algorithm can beadaptively used for denoising photon point cloud data in scenarios with different terrainslopes, different vegetation types, and different acquisition times （daytime and nighttime）.

（2） An improved 2D Progressive Triangular Irregular Network （TIN） Densification（PTD） algorithm was proposed. Based on the fact that ground photons have better slopecontinuity, the study utilized the slope and density information of potential ground pho-tons to reject the residual noisy photons in the lowerpart of the ground， and at the sametime used the density direction information to reject the pseudo ground points broughtabout by misclassification of the lowerpartof the crown vegetation points in the high-density canopy region， so as to realize the accurate extraction of the ground seeds. Finally，the PTD model and the classical sliding window model were utilized for the extractionof ground photons and top-of-canopy photons, respectively. Compared with the ATL08algorithm, the proposed algorithm in this work was able to extract ground photons morecompletely, especially in the region with large slope, and can effectively avoid the in-fluence of residual noise photons in the lower part of the ground on the classification ofphoton point cloud.

（3） Elevation of the ground and the top-of-canopy and canopy height were estimatedfor a variety of ecological sites based on ICESat-2 ATLAS data, and the effects of terrainslope and canopy cover on canopy height were explored. The results show that the con-sistency between the estimated elevation of the ground （MAE = 0.53 m） and the top ofcanopy （MAE = 3.02 m） based on the algorithm proposed in this paper with the referencevalues from the airbore LiDAR scanner is significantly higher than that of the groundelevation （MAE = 1.19 m） and canopy top elevation （MAE = 3.64 m） using the ATL08algorithm. The agreement between the canopy heights of the proposed algorithm andthe corresponding ALS canopy heights is significantly better than that of the ATL08 algorithm results， especially in the daytime data with low signal-to-noise ratio, the accuracy ofthis paper's algorithm improves significantly， with the R' improving from 0.26 to 0.60inATL08. Analysis of the canopy heights error tendency with slope and canopy cover showsthat: the dependence of this paper's algorithm on terrain slope is significantly smaller thanthat of the ATL08 algorithm; ICESat-2 data are more suitable for canopy height inversionin areas with medium to high vegetation cover or temporal phase （≥ 0.4）； and the resultsof canopy height inversion in different biome sites show that ICESat-2 is more suitablefor canopy height inversion for areas covered with wide-expand species.

（4） Based on ATLAS data, the canopy LAI inversion of multiple biome sites wasestimated. The inversion process fully considered the effects of heterogeneous noise den-sity and noise photons inside the canopy on the canopy LAl estimating. The correctionfrom canopy LidAR penetration index to canopy gap fraction was realized by combiningLandsat & OLI multispectral data, and finally the inversion of canopy LAI was realizedusing the gap fraction model. The comparison result with the corresponding ALS LAIshows that ATLAS can be effectively applied to canopy LAI inversion at different eco-logical sites （R = 0.54 ~ 0.94）， and the optimal results are usually obtained at largerscales（~ 100 m）. Comparison with MODIS LAI shows that ATLAS can significantlyimprove the underestimation of LAl by passive optical remote sensing in dense canopies（LAI≥ 4.2）. Overall, ATLAS has good capability for producing global-scale LAI products.