Improving Forest Canopy Cover Mapping Through ICESat-2-Derived Understory DEM and Multisource Remote Sensing Data

Abstract

Canopy cover (CC) is a key indicator of forest horizontal structure. However, in regional-scale forest environments, the spatial heterogeneity of canopy structure, limitations of understory DEMs, and differences in multisource remote sensing responses may still affect the stability and accuracy of CC estimation. This study developed a regional-scale CC estimation framework by integrating ICESat-2 ATL08 samples, random forest-derived digital elevation model (RFDEM), Sentinel-1 C-band and ALOS-2 PALSAR-2 L-band SAR observations, Landsat-9 optical features, forest canopy height (FHC), and land-cover information using a Bayesian optimization-based random forest (BO-RF) model. The framework was evaluated in Georgia, Maryland, and South Carolina and independently validated using G-LiHT airborne LiDAR data. The results show that RFDEM provides a more reliable representation of understory terrain than forest and buildings removed copernicus digital elevation model (FABDEM) in forested areas, with DTM RMSE values of 3.080, 2.308, and 2.221 m in Georgia, Maryland, and South Carolina, respectively, compared with 3.861, 9.507, and 4.478 m for FABDEM. Compared with the FABDEM-based BO-RF models, the RFDEM-based BO-RF models improved test $R^{2}$ and reduced test RMSE in all three study areas. The final BO-RF model achieved $R^{2}$ values of 0.889, 0.893, and 0.892 in Georgia, Maryland, and South Carolina, respectively, during model validation, and maintained $R^{2}$ values of 0.728, 0.796, and 0.755 in independent validation. Feature-combination experiments and SHAP analysis indicate that L-band SAR, FHC, and RFDEM variables provided important information for CC estimation, with their relative importance varying among the three study areas. Overall, the results suggest that combining understory terrain information with optical, SAR, and structural features can improve regional-scale CC estimation under the tested conditions.