针对植被覆盖区TanDEM-X DEM无法描述精细化林下地形的问题，本文提出了一种联合TanDEM-X DEM和Sentinel-2多光谱数据的林下地形提取方法。为了实现该方法，首先将TanDEM-X DEM和Sentinel-2的多波段信息作为输入变量、高精度林下地形数据（LVIS测高数据）作为输出变量，通过随机森林拟合方法构建林下地形预测模型；之后，利用得到的训练模型实现无参考数据区域的林下地形获取。为了验证提出的方法，本文采用位于非洲加蓬的两个典型实验区进行实验验证。实验结果表明，提出的方法能够有效地改正TanDEM-X DEM中包含的森林高度偏差，相较于原始的TanDEM-X DEM，本文方法所提取的地形精度分别提升了76%和63%。此外，实验结果也表明，本文方法得到的林下地形结果保持了较为完整的地形纹理，可以较好的描述林下地形细节。本研究为采用TanDEM-X DEM获取大范围林下地形提供了一种可行的方案。

Digital Elevation Model (DEM) is one of the most important data sources for a variety of scientific studies and applications. Currently, one important data source for large-scale DEM generation origi-nates from the TerraSAR-X add-on for Digital Elevation Measurement (TanDEM-X) mission, which provides bistatic Interferometric Synthetic Aperture Radar (InSAR) data with high spatial resolution (12 m) at the global scale. However, in forest areas, the retrieval of the sub-canopy topography using TanDEM-X InSAR data still faces notable challenges because of the effects of the forest-scattering process on InSAR height measurements and the limited penetration capability of X-band’s signals, which causes the measured elevation to be between the ground surface and the top of the tree canopy. Although SAR signals with longer wavelength has stronger penetrability in the forest layer, sub-canopy topography still cannot be measured due to the volume-scattering effect from tree canopy or trunks. In addition, the missing space-borne PolInSAR or TomoSAR data poses another limitation for the sub-canopy topography estimation. In this paper, a new method to extract sub-canopy topography over forested areas is proposed, which uses a combination of TanDEM-X DEM and Sentinel-2 multispectral data. To achieve this goal, TanDEM-X DEM and the multi-band data of Sentinel-2 are regarded as the input variables, while the high-precision ground elevation data was considered as the target variable; subsequently, random forest fitting method is used to construct the sub-canopy topography estimation predictive model. According to the obtained model, we can extract a large-scale sub-canopy topogra-phy over the areas without reference data. The results show that, evaluated over two different forest sites, in comparison with the original TanDEM-X DEM, the sub-canopy topography derived via the proposed method has an RMSE of 3.7 m and 7.78 m for the two forest sites respectively, which repre-sents an improvement of approximately 76% and 63%, respectively. Furthermore, the experimental re-sults also show that the resultant sub-canopy topography can maintain more detailed topographic in-formation. All these findings indicate that, based on publicly available data, the proposed method has a greater potential for extracting large-scale sub-canopy topography at high spatial resolution.