Forest Height Estimation with Sentinel-1 Dual-Polarization and Interferometric Features Using Random Forest

Forest height is a key structural parameter closely linked to aboveground biomass and ecosystem dynamics. This study evaluates the contribution of Sentinel-1 SAR-derived features, particularly dual polarization and interferometric SAR (InSAR) parameters, in forest height estimation, using Global Ecosystem Dynamics Investigation (GEDI) Light Detection and Ranging (LiDAR) observations as reference. Dual-polarization backscattering coefficients and InSAR coherence and phase features, with or without canopy height models (CHM) derived from GEDI, were used as inputs to a Random Forest regression model. Results show that InSAR features significantly enhance prediction accuracy. The combination of InSAR and CHM achieved a validation $\mathbf{R}^{\mathbf{2}}$ of $\mathbf{0. 6 1 1}(\mathbf{R M S E}=\mathbf{3. 1 1 8 ~ m})$, significantly outperforming the PolSAR + CHM combination, and highlighting the greater predictive power of interferometric features. The full feature set (InSAR + PolSAR + CHM) further improved the performance, yielding the highest accuracy with a validation $\mathbf{R}^{\mathbf{2}}$ of 0.644 and RMSE of 2.884 m. These findings highlight the critical role of InSAR data in improving forest height inversion and underscore its value for integration into large-scale forest monitoring workflows, particularly in the context of advancing remote sensing methodologies for ecological assessment and Earth observation.