ETD

• biomass
• Capon
• forest
• gedi
• lidar
• multiple frequencies
• pct
• polarimetric coherence tomography
• SAR
• tomography
• vertical reflectivity profile reconstruction

The potential of combining polarimetric interferometric (Pol-InSAR) / tomographic (TomoSAR) SAR measurements at multiple frequencies has found successful application in forest mapping. It is widely recognized that, when used in combination with lidar measurements, they enable an improved and/or more complete characterization of the three-dimensional forest structure with metric resolution and repeatedly over time, contributing to the observation and quantitative characterization of forest structure at large scales. Lower frequencies are more sensitive to larger vegetation elements (i.e., on the order of the wavelength). At the same time, the reduced canopy attenuation increases the “visibility” of lower vegetation layers and the ground. Conversely, as the frequency increases, the sensitivity to smaller vegetation elements grows, while the stronger canopy attenuation reduces the “visibility” of the ground. This complementarity in structural information content has recently become particularly relevant in a context where lidar missions such as NASA’s GEDI and ICESat-2, and SAR missions such as ESA’s P-band BIOMASS, NASA’s L-band NISAR, ESA’s L-band ROSE-L, and DLR’s X-band TanDEM-X operate, or will operate, in the same period or in contiguous periods with numerous common objectives related to forest structure, but with different spatial and temporal resolutions and coverages.
The ability to relate the corresponding 3D reflectivities is a critical factor for the development of approaches capable of combining lidar and multi-frequency Pol-InSAR / TomoSAR measurements. Currently, this is generally not established, as no electromagnetic models exist that allow for modeling or transferring reflectivity with the required accuracy across a relevant frequency range. Therefore, in the absence of a model-based method to transfer reflectivity profiles between different frequencies, data-driven solutions remain an attractive — and probably the only truly feasible — option.
In this context, the present thesis explores the potential of polarization coherence tomography (PCT) as a tool to facilitate such transfer. The experimental analysis is based on the processing of real data acquired during the AfriSAR 2016 campaign in the Mondah forest in Gabon. During the campaign, the airborne platforms DLR F-SAR and NASA LVIS almost simultaneously acquired P-band TomoSAR data and lidar waveforms, respectively. In PCT, vertical reflectivity profiles are formulated as a series expansion on a function basis, with coefficients estimated from a limited number of Pol-InSAR measurements. In its original form, PCT uses Legendre polynomials as the basis for reconstructing 3D radar reflectivity, but other types of basis functions can also be adopted, as done in this thesis. The effects of the basis choice are explored in the first part of the work at P-band. Two additional bases have been considered, one derived from P-band TomoSAR profiles (P-band optimized), and one derived from lidar profiles (lidar-optimized). The ability to reconstruct specific profile parameters (for instance the height of the profile local maximum within the canopy and its reflectivity), as well as to reconstruct the InSAR coherences, has been evaluated as a function of the number of used basis components. The experimental results show that a satisfactory reconstruction can be achieved with only a few functions independently of the basis, but the Pol-InSAR geometry is a critical factor affecting the final performance. Clearly, the best reconstruction is obtained using the P-band optimized basis.
In the second part of the work, the focus shifts on the reconstruction of lidar profiles from P-band measurements using a lidar-optimized basis, and the reconstruction of P-band profiles from lidar measurements using a P-band-optimized basis. The experimental results show that in the former case a good approximation of the height of the canopy local maximum can be achieved with just two basis functions, but reconstructing its reflectivity requires a larger dimensionality. In the latter case, reconstructing both height and reflectivity is possible with the first three components, but with larger errors due to the inability to reconstruct unmeasured features because of the lower penetration of lidar pulses compared to P-band ones.
Looking ahead, these findings open promising opportunities. The techniques explored in this thesis—validated on a controlled dataset and a limited test site—offer a concrete starting point for scaling up reflectivity-transfer approaches using operational mission data. This is particularly timely: BIOMASS P-band data are now becoming available, and GEDI products already offer global lidar coverage. Their joint availability will allow immediate large-scale experimentation and validation of the methods proposed here, moving from a few pixels in a single site to continental and global analyses. As such, the approaches developed in this work not only demonstrate feasibility but also pave the way for future multi-mission synergy studies aimed at producing consistent 3D forest structure products across frequencies and sensor types.