森林冠层高度信息对于监测森林的动态变化及定量计算相关的森林生态指标有着十分重要的意义。星载激光雷达作为能够快速、大范围获取森林冠层高度信息的主动遥感数据源，已经成为了森林冠层高度提取的重要方式之一，然而现有的星载激光雷达只能提供离散的沿轨采样数据，所以星载激光雷达冠层测高数据与被动光学影像的结合成为了森林冠层高度制图的常用手段。但以往研究使用的被动光学影像数据往往局限于单角度的观测，对多角度信息在森林冠层高度制图研究中的应用潜力的挖掘比较欠缺，此外，也较少有研究尝试联合多种星载激光雷达数据来参与森林冠层高度制图工作。本文联合ICESat-2、 GEDI两种星载激光雷达数据以及多角度光学影像开展了森林冠层高度制图研究，具体的研究内容及结论如下：

1）利用河北塞罕坝地区的有人机和无人机激光雷达数据对ICESat-2和GEDI的冠层高度产品进行了精度验证及影响因素分析，结果表明：GEDI的产品精度整体优于ICESat-2，在基于有人机数据的验证中，GEDI产品的均方根误差（RMSE）和相关系数（r）分别为3.83 m和0.59，基于无人机数据的验证结果为RMSE=2.57 m和r=0.69，而ICESat-2产品则分别取得了RMSE=4.15 m，r=0.48（有人机）和RMSE=4.46 m，r=0.4（无人机）的验证结果，且两种冠层高度产品均倾向于低估真实冠顶高度。此外，GEDI的产品精度较为稳定，不易受到数据获取时间、植被物候及激光能量等因素的影响，而ICESat-2的产品精度却表现出了一定的波动性，其中，获取于夜间的ICESat-2产品取得了最优的精度验证结果。坡度和森林冠层覆盖度是另外两种能够影响产品精度的重要因素，分析结果表明ICESat-2和GEDI的产品精度均随着坡度的增大而降低、随着森林冠层覆盖度的增大而升高。另外，本文还基于获取于夜间的ICESat-2和GEDI冠层测高数据开展了尺度适用性分析（选取的尺度范围为30~1000 m），分析结果表明这两种冠层测高数据在尺度为30~100 m时能够保持较好且一致的精度，推荐使用最大冠层高度参数参与森林冠层高度制图工作。

2）联合ICESat-2和GEDI的冠层测高数据以及多角度光学影像，基于随机森林回归模型完成了研究区内的森林冠层高度制图。本文利用高分七号、资源三号01星以及PROBA-1 CHRIS影像分别构建了不同的制图场景，同时针对于高分辨率影像对传统的纹理提取方式进行了改进，最终分别制成了31 m、28 m和30 m分辨率的森林冠层高度图并对其进行了精度验证与分析，结果表明：多角度纹理信息对制图结果有重要的影响，在一定条件下，纹理提取方式的改进能够有效提高制图精度（在高分七号的场景中，改进前后场景的RMSE降低了7%，r升高了38%）。高分七号和CHRIS影像场景的制图结果与有人机激光雷达数据的第95百分位高度的相关性最好，分别取得了RMSE=2.49 m，r=0.36、和RMSE=2.96 m，r=0.58的精度验证结果，资源三号01星与第97百分位高度的相关性最好，取得了RMSE=2.85 m，r=0.43的验证结果。高分七号和资源三号01星影像场景的制图精度在很大程度上受到了影像数据获取自落叶期的限制，因此，联合获取于有叶期的高分辨率多角度光学影像与星载激光雷达数据有望使得森林冠层高度制图精度得到进一步的提升。

Forest canopy height is of great significance for monitoring the dynamic changes of forests and quantitatively calculating relevant ecological indicators. Spaceborne lidar, as an active remote sensing technology that can directly detect the forest canopy height at large scales, has become one of the important ways to extract forest canopy height, however, existing spaceborne lidars can only provide discrete along-track measurements, so the combination of spaceborne lidar derived canopy heights and passive optical images has become a common method for forest canopy height mapping. But the passive optical images used in previous studies are often limited to single-angle observations, and the application potential of multi-angle information in forest canopy height mapping is relatively lacking, in addition, few studies have attempted to use multiple spaceborne lidar datasets in forest canopy height mapping research. This paper combines two spaceborne lidars which are ICESat-2 and GEDI and multi-angle optical images to map forest canopy height of Saihanba, Hebei Province, the specific research contents and conclusions are as follows:

1) Accuracy assessments and influential factors analysis of canopy height products from ICESat-2 and GEDI using airborne and UAV lidar datasets were carried out, results show that GEDI outperforms ICESat-2 in both airborne (RMSE=3.83 m, r=0.59; vs. RMSE=4.15 m, r=0.48) and UAV (RMSE=2.57 m, r=0.69; vs. RMSE=4.46 m, r=0.4) lidar-based assessments, and they both underestimate canopy top heights. GEDI is insensitive to data acquisition time, vegetation phenology and laser energy, while nighttime data are recommended for ICESat-2 users. Slope and forest canopy coverage are two other important factors that can affect the product accuracy, results show that the product accuracy of ICESat-2 and GEDI both decreased with the increase of slope and increased with the increase of forest canopy coverage. In addition, an evaluation of ICESat-2 (acquired at night) and GEDI canopy height measurements’ applicability at different scales was also conducted (the scale ranges from 30 to 1000 m), and results show that the two kinds of canopy height data can maintain good and consistent accuracy when the scale is 30~100 m, and it is recommended to use the maximum canopy height in forest canopy height mapping research.

2) Combined with the above-mentioned two kinds of canopy height data and multi-angle optical images, the forest canopy height in the study area was mapped based on the random forest regression model. In this paper, different mapping scenes were constructed using GF-7, ZY-3 01 and PROBA-1 CHRIS images, the traditional texture extraction method was also revised for the coarse-scale texture calculation of high-resolution images, and finally, the forest canopy height maps with resolutions of 31 m, 28 m and 30 m were produced and their accuracies were verified and analyzed respectively. Results show that the multi-angle texture information has a great influence on mapping results, and under certain conditions, the revised texture extraction method can effectively improve the mapping accuracy (in the GF-7 scene, RMSE decreased by 7% and r increased by 38% after using the revised method). The mapping results of GF-7 and CHRIS image scenes have the best correlation with the 95th percentile canopy height of airborne lidar data, while ZY-3 01 has the best correlation with the 97th percentile canopy height, accuracy verification results of RMSE=2.49 m, r=0.36, RMSE=2.96 m, r=0.58 and RMSE=2.85 m, r=0.43 were obtained respectively. The mapping accuracy of GF-7 and ZY-3 01 satellite image scenes is largely limited by the image datasets acquired during the deciduous period, therefore, the combination of high-resolution multi-angle optical images acquired at leaf-on seasons and spaceborne lidar data is expected to further improve the accuracy of forest canopy height mapping.