Quantitative Interpretation of the Penetration Process of L-Band SAR Signals in Mapping Chinese Fir Forest Canopy Heights With Quad-Polarization SAOCOM-1 Images

The penetrating characteristics of Synthetic Aperture Radar (SAR) signals are widely recognized as a key advantage over optical remote sensing data. However, this penetration capability can lead to an underestimation of forest canopy height when InSAR or PolInSAR techniques are used, particularly with L-band SAR signals. Due to the complex internal structure of forests and the strong penetration ability of L-band SAR signals, quantitatively characterizing how these signals interact with forest canopies remains a significant challenge. To address the limitations imposed by SAR signal penetration on mapping forest canopy height, simulated L-band quad polarization data and spaceborne SAOCOM-1 images were utilized to analyze the interaction mechanisms between L-band SAR signals and forest canopies with InSAR technology. Subsequently, two quantitative parameters, Penetration Depth Height (PDH) and Penetration Depth Rate (PDR), were introduced to measure penetration depth and examine its variation across different forest types and polarization channels. Finally, by evaluating the relationships between these proposed parameters and actual forest heights, a novel forest height mapping approach was proposed by integrating the XGBoost model with PDH-related variables derived from SAOCOM-1 imagery, with the assistance of GF-7 stereo images. The results indicated that the penetration depth of L-band SAR signals in forests is strongly influenced by tree species, forest height, and canopy density. Coniferous forests exhibited significantly greater penetration capability compared to broad-leaved forests. Moreover, the PDH across the three polarization channels followed a consistent order from shallow to deep: HV, VV, and HH. With increasing forest height and density, the PDR in coniferous forests gradually decreased to 0.5, corresponding to the lower canopy layers, whereas in broad-leaved forests, the penetration depth was predominantly observed in the upper half of the canopy. Among the quantified parameters, PDH demonstrated the strongest correlation with forest height. The results also demonstrated that the proposed method for forest height mapping yielded promising performance, achieving root mean square error (RMSE) values of 3.86 m and 5.01 m, with corresponding relative RMSE (rRMSE) values of 21.7% and 36.2% for coniferous and broad-leaved forests, respectively. These findings suggest that the proposed approach has considerable potential for large-area forest height mapping.