Characterization of forests structure from interferometric synthetic aperture radar

Large-scale forest monitoring is essential for understanding ecosystem dynamics, quantifying carbon stocks, and supporting climate change mitigation and sustainable management efforts. Spaceborne Synthetic Aperture Radars (SAR) provide consistent and systematic observations of forests, offering global coverage and short revisit times. The growing availability of multi-temporal and multi-frequency SAR datasets presents new opportunities for accurate and consistent large-scale mapping of forest structural parameters. This dissertation investigates the use of Interferometric SAR (InSAR) data for retrieving forest structure, with the aim of improving the understanding of radar–forest interactions and developing scalable methods for tree height estimation from spaceborne and airborne InSAR acquisitions. The first part of the work focuses on the physical modeling of temporal decorrelation over forests and the development of model-based frameworks for tree height retrieval from multi-temporal backscatter and interferometric coherence data. Two retrieval algorithms, specifically designed for high- and low-frequency datasets, are proposed and tested using C-band Sentinel-1 and L-band ALOS-2 interferometric time series. The methods rely on joint physical modeling of radar backscatter and interferometric coherence over forests and explicitly account for the influence of varying environmental conditions on the radar observables. In addition to the model-based analysis, the thesis investigates a data-driven approach based on deep learning to assess the capability of neural networks to represent the variability of interferometric radar observables and their relationship with forest parameters. Convolutional neural networks are trained to learn the complex relationships between radar observables and tree height, and their performance is evaluated using L-band ALOS-2 interferometric time series and reference LiDAR measurements. The final part of the dissertation explores the potential of SAR Histomography for mapping forest vertical structure. This recently introduced technique emulates tomographic capabilities while relying on a single interferometric pair, providing an approximation of tomographic-like profiles with reduced acquisition requirements. The analysis is performed using L- and P-band airborne UAVSAR data acquired over tropical forests in Gabon. Structural and statistical metrics are introduced to assess the variability of radar-derived forest profiles as a function of system parameters and scene characteristics. Furthermore, a preliminary algorithm is proposed to estimate digital elevation models and forest height from histomography-derived profiles while accounting for signal attenuation through the canopy. Overall, this work advances the understanding and use of interferometric radar data for forest applications, providing both theoretical and practical insights toward reliable and scalable approaches for forest structure mapping.