Hierarchically Optimized Forest Coverage Modeling With Multisource Remote Sensing Data for High-Precision Understory Terrain Mapping

The uncertainty in the accuracy of understory terrain severely restricts its application in scientific research, while precise terrain information is crucial for studies such as vegetation biomass estimation. In this paper, an innovative data fusion method is proposed, which integrates multi-source remote sensing data and adopts a forest coverage-based hierarchical optimization modeling strategy to construct a machine learning(ML) model for understory terrain estimation. Specifically, we utilize multi-modal spaceborne LiDAR data (GEDI and ICESat 2), Sentinel data, high-precision digital elevation models(DEMs), forest structural parameters, and topographic factors to build ML models following the forest cover hierarchical strategy. The experimental results show that among the evaluated ML algorithms, the Random Forest (RF) model achieves the optimal performance, improving the accuracy of understory terrain by 51% at the regional scale compared with FABDEM(RMSE reduction from 6.78m to 3.30m). In addition, the hierarchical modeling based on forest cover further enhances the accuracy by 13% (to 2.85m RMSE)in Oregon study area. The study also reveals that the accuracy of understory terrain is affected by forest types and soil properties; particularly, forest types with dense canopies and complex structures, as well as loose-textured soils susceptible to vegetation interference, exert a significant impact on the accuracy.