Conference Agenda

Polarimetric SAR Interferometry (PolInSAR) merges polarimetric SAR, offering the ability to distinguish between different scattering types within individual resolution cells, and interferometric SAR, which provides sensitivity to the vertical stratification of scatterers.

The PolInSAR two layer model represents the signal as a combination of the ground and the volume contributions. It has been especially useful to model the PolInSAR signal over vegetation. Taking advantage of the fact that these two components have different polarimetric and interferometric characteristics, it is possible to extract physical parameters such as the ground phase or the height of the forest [1]. Recently, this model has also been used to separate and extract fully polarimetric signatures from the ground and volume, as a polarimetric decomposition [2].

When several baselines are available, the vertical reflectivity profile of the vegetation can be estimated. This is equivalent to a synthetic aperture in the across-track dimension, as done in SAR Tomography. In this sense, the vertical reflectivity profile estimation is converted into a spectral estimation. Assuming the PolInSAR two layer model, the vertical reflectivity of the ground and volume components define their interferometric coherences, linking polarimetric and interferometric dimensions of the data [3]. In this work, we propose exploiting the spatial variability of the volume layer height across the image, instead of requiring additional baselines, to estimate its profile. The additional assumption then is that the volume profile of spatially close stands may be seen as the same normalized profile scaled in height.

The proposed technique has been evaluated with airborne data from the DLR FSAR sensor over the AfriSAR campaign at P-band, providing a reasonable volume profile over tropical forests when using a single baseline. This technique could be very relevant for the BIOMASS mission, where only a limited number of baselines will be available during the interferometric phase, allowing the analysis of the forest vertical profiles and its changes over time.

References

[1] S. Cloude and K. Papathanassiou, “Three-stage inversion process for polarimetric SAR interferometry,” IEE Proceedings-Radar, Sonar and Navigation, vol. 150, no. 3, pp. 125–134, 2003.

[2] A. Alonso-Gonzalez and K. P. Papathanassiou, “Multibaseline two layer model polinsar ground and volume separation,” in EUSAR 2018; 12th European Conference on Synthetic Aperture Radar, 2018, pp. 1–5.

[3] A. Alonso-González, N. Romero-Puig, C. López-Martínez, and K. P. Papathanassiou, “Polarimetric Ground and Volume Decomposition by means of PolInSAR two layer model,” Submitted to IEEE Transactions on Geoscience and Remote Sensing, under review.

Advanced spaceborne 3D Tomographic SAR (Tomo-SAR) [1-5] missions based on companion/tandem (i.e. formation-flying) configurations are emerging for remote sensing of forests [2-4,6], key in the CO2 cycle matters. In fact, this implementation of Tomo-SAR, obtaining the height profile from a sequence of track-pair only coherence data, bypasses typical long-term decorrelation issues of repeat pass monostatic SAR configurations [2,3]. However, because of safety/orbit control issues, the formation satellites do not pass exactly simultaneously [6,7]. Therefore, wind-induced short-term (fractions of second scale) decorrelation phenomena can still affect tandem Tomo-SAR processing [7] in forest scenarios. Also, it is noted that the Tomographic blurring sources are height-localized [2,3].

In our work, a methodology is thus reported for new analyses of 4D (3D + Time [5]) Differential Tomography type of short-term forest temporal decorrelation phenomena, exploiting a ground-based X-band miniradar array (ack. IDS Italy), placed vertically, with very quick acquisition capabilities (up to 1000 array firings per second) [7]. In particular, multiple vertical beamformings are performed, one for each array snapshot, followed by height-by-height statistical analyses along the quickly sampled temporal domain.

Moreover, corresponding real data results have been obtained, developing characterizations of both height- and time-varying behaviours of short-term decorrelation in a representative experiment for a stand of elms and poplars trees in Italy (UniPi PisaScat experiment [7]). The data were acquired in two different wind conditions and seasons.

For example, advanced height-varying Doppler spectra estimates have been obtained; increasing Doppler bandwidths with height and for stronger wind velocity resulted, indicating corresponding increasing velocity of the random motions of windblown branches and leaves.

Through the well-known signal bandwidth-coherence time relation, short-term coherence times (for coherence to decay to a steady-state) have been also measured; these can be useful to state if the short-term decorrelation issue influences a Tomo-SAR tandem satellite system, according to its formation-flying time-lag [6], or if this issue can be neglected. Noteworthy, a variability of coherence time along the trees layers (decreasing with higher heights), and a decorrelation timescale of small fractions of second, have resulted to be apparent. As a comparison, ensemble (height-integrated) Doppler measures have also been carried out, resulting mostly sensitive just to the lower (less critical) heights.

Other obtained findings include height-varying short-term steady-state coherence levels (which can tell about influence degree for tandem satellite Tomo-SAR); time-varying height scattering profiles and (sliding window-based) mean Doppler shifts under wind gusts, for the different heights; and direct autocorrelation function measures of the scattering at very short sampling scale (representing a physical view of the height-varying short term decorrelation process and a confirmation of the Doppler-based coherence time results). These will be presented at the conference, together with physical comments, and other acquisition system, scenario and methodology details.

Such investigations of height-varying characteristics of short-term decorrelation processes complement ESA radar tower [3] measures, which are instead at some seconds scale and basically of ensemble kind.

Our advanced characterization methodology and the obtained findings (possibly carrier frequency-scaled) may be useful for development, or definition refinement of the operative envelope, of the emerging tandem satellite programs and missions; examples are LuTan-1, already operative (as TanDEM-X), the NASA DART, and SAOCOM-CS-like system [6] concepts, currently being developed.

Ack.: This work has been partially supported by the Italian Ministry of Education and Research (MUR) in the framework of the FoReLab project (Departments of Excellence).

[1] T.G. Yitayew, L. Ferro-Famil, T. Eltoft, “High Resolution Three dimensional Imaging of Sea Ice using Ground-based Tomographic SAR Data,” Proc. 10th EUSAR, Berlin, Germany, 2014, pp.1325-1328.

[2] M. Lavalle, M. Simard, S. Hensley, “A Temporal Decorrelation Model for Polarimetric Radar Interferometers,” IEEE TGRS, 50(7), pp.2880-2888, 2012.

[3] T. Dinh Ho Minh, et al., “Vertical Structure of P-Band Temporal Decorrelation at the Paracou Forest: Results from TropiScat,” IEEE GRSL, 11(8), pp.1438-1442, 2014.

[4] A. Reigber, A. Moreira, “First Demonstration of Airborne SAR Tomography using Multibaseline L-band Data,” IEEE TGRS, 38(5), pp.2142-2152, 2000.

[5] D. Reale, G. Fornaro, A. Pauciullo, X. Zhu, R. Bamler, “Tomographic Imaging and Monitoring of Buildings with Very High Resolution SAR Data,” IEEE GRSL, 8(4), pp.661-665, 2011.

[6] P. López-Dekker, H. Rott, P. Prats-Iraola, B. Chapron, K. Scipal and E. D. Witte, “Harmony: an Earth Explorer 10 Mission Candidate to Observe Land, Ice, and Ocean Surface Dynamics,” Proc. 2019 IEEE IGARSS, Yokohama, Japan, 2019, pp.8381-8384.

[7] F. Lombardini, G. Buttitta, “On the Issue of Short-term Decorrelation in Forest Tomography with Companion SARs,” Proc. 2024 IEEE IGARSS, Athens, Greece, 2024, pp.3104-3107.

Forest SAR tomography is a key tool for monitoring biomass. It consists of processing a stack of coherent SAR images to retrieve several observables related to the biomass. One of the challenges of forest SAR tomography is that the shape of the forest’s reflectivity profile is unknown. Usually, some prior is introduced, e.g., the reflectivity profile is assumed to have a Gaussian or an exponential shape. However, the choice of this prior impacts the estimation performance, as an incorrect model induces estimation biases.

This communication focuses on estimating the mean height, the spread, and the total backscattered power of the forest reflectivity profile. It presents a new estimation method that does not require, nor assume, any shape for the reflectivity profile of the forest. Instead, the unknown reflectivity distribution is characterized by its central moments.

First, a discussion on the observability of the forest SAR tomography problem justifies the introduction of the moments to characterize the distribution of scatterers. The key idea is that the spread of the scatterers is small with respect to the system ambiguity, typically 5 meters vs. 90 meters. Therefore, the spectrum of the reflectivity density is smooth and can be approximated by the first terms of its Taylor expansion, which involves the moments of the distribution.

Based on this observation, a new model for the covariance matrix is obtained by cutting the Taylor expansion at some order D, which is a hyperparameter to set. In this model, the first D moments are (new) parameters to estimate, along with the usual parameters. Then, a new estimation algorithm is proposed by applying Covariance Matching Estimation Techniques (COMET) to fit the empirical covariance of the measurements with the proposed model using a least-squares minimization. The main advantage of this model is its computational complexity. All the parameters but one, the mean height of the forest, enter linearly in the model of the covariance matrix. Consequently, the estimation algorithm requires only the optimization of a single parameter. Moreover, as the model relies on the moments and does not assume any prior on the shape of the reflectivity profile, it is robust to incorrect modeling. However, the cut-off induces a misspecification and potentially estimation biases.

The performance of the algorithm is assessed in realistic SAR tomography simulations. Scenarios are proposed to estimate the forest profile with and without ground cancellation. The estimation algorithm is compared with the maximal-likelihood estimator (MLE), which assumes a shaping distribution. As expected, when well-specified, the MLE obtained the best performance, but it is outperformed by the proposed algorithm when the assumed shape is incorrect. Finally, the choice of the hyperparameter is discussed and a tradeoff is highlighted. Large cutting orders allow better reconstructions with smaller biases but are slower to converge, while small cutting orders are more robust but present larger biases.

Due to its robustness and its low complexity, the proposed algorithm is particularly suited for forest SAR tomography, where it can efficiently process wide areas. The next step will be to apply it to Biomass data.

This study aims to characterize forest structure using multi-baseline Synthetic Aperture Radar (SAR) tomography, leveraging ESA’s Biomass Tomography Mission P-band SAR products for calibration and validation (Cal/Val). The study focuses on retrieving 3D forest structure, estimating forest height, and analyzing biomass distribution using advanced SAR tomographic techniques. The Biomass Tomography Mission provides P-band SAR data specifically designed for forest parameter estimation, offering deep penetration capabilities to extract vertical structure information.

Materials:

For this study, we have chosen two forest sites over Gabon which is taken in descending pass and over Amazon forest which is taken in both ascending and descending passes from the 32 datasets that are provided. We will use Level 1A - SCS products for this study. We intended to use only June GEDI sample points, but currently, it is not published for June 2025. Hence, GEDI LiDAR data from Jan 01 2025 till 02 May 2025 over the two sites is considered.

Methodology:

We will employ multi-baseline SAR tomography techniques, including spectral estimation methods such as Beamformer, Capon, MUSIC, and Compressive Sensing. These techniques enable the separation of different scattering contributions within forested areas, leading to a more accurate representation of vertical forest structure. The processing will include interferometric phase calibration, coherent combination of SAR acquisitions, and tomographic inversion for reconstructing vertical reflectivity profiles.

The Biomass Tomography Mission's P-band SAR products, particularly tomographic forest height and biomass estimates, will be used as reference datasets to validate our reconstructions. The calibration and validation process will involve comparison with ground truth data, including GEDI LiDAR-derived forest height measurements, in-situ biomass inventories, and other independent datasets. Additionally, error estimation and uncertainty quantification will be performed to assess the robustness of the tomographic inversion results.

Expected Outcomes:

High-resolution 3D forest structure maps: Using multi-baseline SAR tomography, we will generate spatially detailed representations of forest height and biomass distribution. Improved biomass estimation models: By incorporating validated tomographic inversion results, we aim to refine biomass retrieval algorithms and enhance their accuracy.

Comparative assessment with Biomass Tomography Mission products: A detailed evaluation of our tomographic retrievals against ESA’s Biomass Mission P-band SAR products and also GEDI LiDAR retrievals will be conducted to assess the consistency and accuracy.

Uncertainty analysis: A comprehensive error assessment will be carried out to quantify uncertainties in tomographic height and biomass estimates.

Datasets and reports: The results of this study, including tomographic forest structure datasets, will be available for further research and applications in global biomass mapping.

Synthetic Aperture Radar (SAR) tomography (TomoSAR) exploits multi-baseline acquisitions coherently combined to estimate a three-dimensional reflectivity distribution of the scene. Classical reconstruction methods, however, strongly depend on missions with a large number of tracks to achieve high-quality tomograms in terms of resolution and effective ambiguity suppression. This challenge is particularly pronounced in forested regions, where scattering contributions from vegetation volume and ground surfaces overlap along the reconstruction direction, increasing the dependence of the tomographic result on the stack size.

We propose using a U-Net, a deep learning architecture designed for dense prediction tasks, to directly map reduced TomoSAR stacks into high-quality vertical profiles, comparable to those obtained with significantly larger baseline stacks using classical methods.

The reference data is obtained from Capon vertical profiles computed using a highly dense stack. We further leverage a model of the expected vertical backscattering reflectivity distribution (defined by a set of physical parameters) to incorporate physical prior knowledge into the reference generation, also helping the proposed network to fit the data.

The proposed method is evaluated using L-band tomographic data acquired by DLR’s F-SAR system over the Traunstein temperate forest site in southern Germany in 2017, demonstrating its potential to reduce acquisition requirements to one-fifth of the original amount while maintaining reconstruction quality. The reconstruction quality is evaluated against high-quality profiles using quantitative similarity measures applied both to the model parameters and to the reconstructed tomographic profiles.