Up-to-date high-resolution understory terrain extraction based on satellite stereo photogrammetry and spaceborne LiDAR

Accurate digital terrain models (DTMs) are essential for a wide range of geospatial and environmental applications, yet their derivation in forested regions remains a significant challenge. Existing global DTMs, typically generated from satellite stereo photogrammetry or interferometric synthetic aperture radar (InSAR), fail to accurately capture understory terrain due to limited penetration capabilities, resulting in elevation overestimation in densely vegetated areas. While airborne light detection and ranging (LiDAR) can provide high-accuracy DTMs, its limited spatial coverage and high acquisition cost hinder large-scale applications. Thus, there is an urgent need for a scalable and cost-effective approach to extract DTMs directly from satellite-derived digital surface models (DSMs).In this study, we propose a simple, interpretable understory terrain extraction method that utilizes canopy height data from Global Ecosystem Dynamics Investigation (GEDI) and Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) to construct a tree height surface model, which is then subtracted from the stereo-derived DSM to generate the final DTM. By directly incorporating LiDAR constraints, the method avoids error propagation from multiple heterogeneous datasets and reduces reliance on ancillary inputs, ensuring ease of implementation and broad applicability. In contrast to machine learning-based terrain modeling methods, which are often prone to overfitting and data bias, the proposed approach is simple, interpretable, and robust across diverse forested landscapes. The accuracy of the resulting DTM was validated against airborne LiDAR reference data and compared with both the Copernicus Digital Elevation Model (DEM) and the forest and buildings removed DEM (FABDEM), a global bare-earth elevation model corrected for vegetation bias. The results indicate that the proposed DTM consistently outperforms the Copernicus DEM (CopDEM) and achieves accuracy comparable to FABDEM. In addition, its finer spatial resolution of 1 m, compared to the 30 m resolution of FABDEM, allows for more detailed terrain representation and better capture of fine-scale variation. This advantage is most pronounced in gently to moderately sloped areas, where the proposed DTM shows clearly higher accuracy than both the CopDEM and FABDEM. The results confirm that high-resolution DTMs can be effectively extracted from DSMs using spaceborne LiDAR constraints, offering a scalable solution for terrain modeling in forested environments where airborne LiDAR is unavailable.To illustrate the potential utility of the proposed DTM, we applied it to a fire risk mapping application based on topographic parameters such as slope, aspect, and elevation. This case highlights how improved terrain representation can support geospatial hazard assessments.