Assessing the Two-Layer Forest Scattering Model Using Sentinel-1 Dual-Polarimetric InSAR Observations

Abstract

Physics-based two-layer polarimetric interferometric models provide a well-established framework for forest parameter retrieval. Yet, under Copernicus Sentinel-1 C-band repeat-pass acquisition constraints, the empirical behavior and characteristics of the random motion over ground (RMoG) model parameters remain largely unexplored. To gain insights into the behavior of the key model parameters—including the ground-to-volume scattering ratio, volumetric attenuation, vegetation motion decorrelation, and forest height—we assemble a large, globally distributed benchmark dataset by combining Sentinel-1 dual-polarimetric, single-baseline interferometric coherence observations with independent forest height measurements from GEDI and ICESat-2. The dataset spans boreal, temperate, and tropical forest biomes, providing a comprehensive basis for the large-scale analysis of the RMoG model parameters under real acquisition conditions. The analysis reveals distinct empirical behaviors among the RMoG parameters. Vegetation motion decorrelation exhibits a strong and consistent relationship with interferometric coherence and forest height across all investigated forest biomes. In contrast, volumetric attenuation and the ground-to-volume scattering ratio exhibit high variability and broad distributions across the coherence range. These findings motivate a simplified RMoG inversion through the direct parameterization of vegetation motion decorrelation, reducing the inversion dimensionality while maintaining a physically based estimation framework. Furthermore, the assembled synthetic aperture radar (SAR)–LiDAR dataset motivates the investigation of machine learning as a means of compensating for the limitations of the physical model, yielding further improvements in forest height estimation. Experimental results demonstrate that, despite the challenging Sentinel-1 C-band repeat-pass acquisition conditions, the physics-based RMoG inversion preserves the dominant spatial variability of forest height and remains suitable for discriminating broad forest height classes.