Conference Agenda

Spaceborne SAR systems represent a powerful mean for monitoring forest on a global scale. In particular, time-series data provided by the Sentinel 1 C-band SAR mission, have been widely used for detecting deforestation [1,2], or measuring forest degradation [3]. SAR sensors operating at L or P bands use larger-wavelength signals that can penetrate dense forest canopies to the ground, and can be used to retrieve certain geophysical forest features, such as above-ground biomass [4,5]. However, the robust and accurate estimation of internal forest descriptors is usually hampered by multiple wave-matter interactions that occur at the ground level and contribute significantly to the total SAR response. These undesired components exhibit highly variable radiometric and polarimetric patterns, influenced by numerous factors, such as acquisition geometry, local topography, soil humidity and roughness...

SAR tomography constitutes a solution for discriminating echos from a forest canopy and the ground [6] : multiple coherent signals acquired from slightly offset trajectories are focused in 3D, providing access to the reflectivity of forest components located at different elevations. Nevertheless, vertical separation is, in general, not perfect, and an unrealistic number of SAR acquisitions may be required to achieve a sufficient level of isolation between the imaged responses of the ground and the overlying volume. Another approach consists in canceling out responses originating from the ground level by coherently combining a pair of SAR images [7]: the intensity of the resulting image is a non-linear function of the above-ground reflectivity of the scene, and depends on multiple, often unknown, factors, such as acquisition geometry, local topography, forest structure…

This paper proposes generating 3D reflectivity maps that are insensitive to ground scattering by generalizing the coherent ground filtering principle introduced in [7] to the case of SAR tomography.

This combined processing cancels out the undesired component with a level of isolation that does not depend on the vertical tomographic resolution, while yielding a refined image of the forest 3D reflectivity. The method is based on an unconstrained optimization problem whose analytical solution may be applied to non-parametric tomographic focusing, e.g. Beamformer or Capon’s method, of single- or multi-look SAR data. The techniques can handle polarimetric SAR data and deliver optimal or full-rank ground-notched 3D polarimetric information.

The performance of the approach is assessed through a thorough comparison with the aforementioned methods, using measurements from ESA’s airbone SAR campaigns and early BIOMASS data.

[1] Reiche, J., Verhoeven, R., Verbesselt, J. et al. Characterizing Tropical Forest Cover Loss Using Dense Sentinel-1 Data and Active Fire Alerts. Remote Sensing, 10, 777 (2018).

[2] Bottani, M., Ferro-Famil, L., Doblas, J. et al. Novel unsupervised Bayesian method for Near Real-Time forest loss detection using Sentinel-1 SAR time series: Assessment over sampled deforestation events in Amazonia and the Cerrado. Remote Sensing of Environment, 331, 115037 (2025).

[3] Dupuis, C., Fayolle, A., Bastin, J.F. et al. Monitoring selective logging intensities in central Africa with sentinel-1: A canopy disturbance experiment. Remote Sensing of Environment, 298, 113828 (2023).

[4] Le Toan, T., Quegan, S., Davidson, M. W. J. et al. The BIOMASS mission: Mapping global forest biomass to better understand the terrestrial carbon cycle. RSE, 115(11), 2850-2860. (2011).

[5] Bouvet, A., Mermoz, S., Le Toan, et. al. An above-ground biomass map of African savannahs and woodlands at 25 m resolution derived from ALOS PALSAR. Remote sensing of environment, 206, 156-173. (2018)

[6] Aghababaei, H., Ferraioli, G., Ferro-Famil, L. et. al.. Forest SAR tomography: Principles and applications. IEEE geoscience and remote sensing magazine, 8(2), 30-45 (2020).

[7] Mariotti d’Alessandro, M. Tebaldini, S., Quegan et. al. Interferometric ground cancellation for above ground biomass estimation. IEEE Transactions on Geoscience and Remote Sensing, 58(9), 6410-6419. (2020)

ESA’s BIOMASS, was launched on 29th of April 2025 to capture interferometric polarimetric and tomographic information, for the first time enabling the P-band spaceborne Pol-InSAR forest height inversion [1]. The spatial baseline distribution of BIOMASS is optimized for tropical forest height retrieval and tomographic profile reconstruction, while the satellite’s short revisit time of three days provides a unique opportunity to obtain high-quality forest height maps [2].

It is well known that forest height correlates with cross-track interferometric coherence [2]. However, obtaining high-quality P-band interferograms remains challenging due to ionospheric propagation effects, correct accounting of temporal decorrelation, dielectric change due to mainly rain events and correct approximation of vertical reflectivity profile [3-4]. This paper addresses these challenges step by step, leading to the implementation and validation of a forest height inversion algorithm. The first results demonstrate the immense potential of BIOMASS in delivering high-quality data and accurate forest height measurements.

The different model-based strategies of forest height from provided BIOMASS data will be discussed: single-, dual- and three-baseline forest height inversion scenarios. The single-baseline inversion approach is straightforward in interpretation and computationally efficient; however, it requires prior knowledge of the ground phase and assumptions regarding temporal decorrelation. In contrast, multi-baseline inversions (dual- and three-baseline) enable a more balanced estimation framework but may lead to ambiguous solutions and increased uncertainty in result interpretation [3].

This study will provide a full analysis of forest height inversion potential with BIOMASS, using its full dataset and addressing the inversion in physics-aware model-based framework. The primary focus is on assessing performance across the rainforests of the Amazon, Central Africa, and Southeast Asia. Results will be closely linked to the mission’s commissioning phase and critically validated using GEDI’s LiDAR-derived RH98 forest height data, ensuring the verification of BIOMASS-derived height products.

REFERENCES

[1] S. Quegan et al., The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space. Remote Sensing of Environment, 227, 44-60, 2019

[2] S. R. Cloude and K. P. Papathanassiou, "Polarimetric SAR interferometry," in IEEE Transactions on Geoscience and Remote Sensing, vol. 36, no. 5, pp. 1551-1565, Sept. 1998, doi: 10.1109/36.718859.

[3] R. Guliaev, J. Su Kim, M. Pardini and K. P. Papathanassiou, "On the Use of Tomographically Derived Reflectivity Profiles for Pol-InSAR Forest Height Inversion in the Context of the BIOMASS Mission," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-12, 2024,

[4] M. Pardini, M. Tello, V. Cazcarra-Bes, K. P. Papathanassiou and I. Hajnsek, "L- and P-Band 3-D SAR Reflectivity Profiles Versus Lidar Waveforms: The AfriSAR Case," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 10, pp. 3386-3401, Oct. 2018, doi: 10.1109/JSTARS.2018.2847033

Monitoring forest structure is a key objective of the recently launched ESA’s BIOMASS mission [1]. It operates at P-band (70 cm wavelength) to retrieve forest height, biomass, and disturbance at global scale. The long wavelength provides sensitivity to the full canopy depth, offering unique opportunities to characterize forest scattering mechanisms from canopy to ground. This study presents first analyses of forest structure mapping using early BIOMASS P-band fully polarimetric and interferometric data.

Natural forest scenes are complex radar scattering environments, where surface, dihedral, and volume contributions coexist within a single resolution cell. While polarimetric SAR (PolSAR) techniques can reveal scattering diversity, they cannot resolve its vertical distribution. Conversely, interferometric SAR (InSAR) provides height sensitivity but is influenced by temporal and environmental decorrelation. The combination of both by means of Pol-InSAR [2] techniques offers the potential to jointly exploit scattering diversity and height information for structural mapping.

To this end, this work proposes to first separate the fully polarimetric response into simpler scattering contributions by means of interferometry, i.e., ground and volume [3]. The ground response is then modelled as a mixture of elementary surface (Bragg) and dihedral (Fresnel) mechanisms, simplifying the interpretation of their polarimetric signatures [4]. Preliminary results using fully polarimetric COM-1 acquisitions already demonstrate the excellent radiometric and polarimetric quality of the BIOMASS data. The scattering entropy can be employed as a measure of polarimetric coherence and spatial variability. The proposed decomposition effectively distinguishes forest areas with open and closed canopies, highlighting the mission’s sensitivity to structural diversity even prior to full calibration.

These results contribute to the ESA’s BioTomEx project and support the development of BIOMASS-derived forest disturbance products. Data from the GABON-X campaign that will take place in November 2025 will help consolidating proposed Pol-InSAR approaches for forest structure monitoring by comparison with BIOMASS data.

[1] S. Quegan et al., “The European Space Agency BIOMASS mission: Measuring Forest above-ground biomass from space,” Rem. Sens. Environment, vol. 227, pp. 44-60, June 2019.

[2] K. P. Papathanassiou and S. R. Cloude, “Single-baseline polarimetric SAR interferometry,” IEEE Trans. Geosci. and Remote Sens., vol. 39, no. 11, pp. 2352-2363, Nov. 2001.

[3] Alonso-González, A., Papathanassiou, K. “Polarimetric Change Analysis in Forest using PolInSAR Ground and Volume separation techniques,” EUSAR 2021; 13th European Conference on Synthetic Aperture Radar, pp. 1-6. VDE, 2021.

[4] A. Freeman and S. L. Durden, “A three-component scattering model for polarimetric SAR data,” IEEE Trans. Geosci. and Remote Sens., vol. 36, no. 3, pp. 963-973, May 1998.

ESA BIOMASS mission, operating in orbit since April 2025, is designed to provide near global estimates of forest biomass, structure and change. In this study, we present a first evaluation of the ESA BIOMASS mission Level-1B In-Orbit Commissioning (IOC) data over Brazilian forests covering a wide range of biomass, from dense tropical forests to savannas. We compared BIOMASS amplitude data with independent reference data from the Brazilian National Forest Inventory (NFI) and INPE Airborne Laser Scanning (ALS) data. Furthermore, we superimposed BIOMASS data with a secondary forest age product, to determine if there was a relationship between BIOMASS signal and forest age. Although the L1B IOC data are not yet radiometrically or polarimetrically calibrated, the results show promising relationships between the BIOMASS amplitude and key forest structural parameters (aboveground biomass and canopy height) as well as forest age.

As expected, the strongest correlations with forest structure were observed at cross-polarizations (HV and VH). Moreover, the cross-polarized P-band signal exhibits a strong relationship with the estimated age of secondary forests. Finally, the uncalibrated BIOMASS data shows stronger correlations with forest structure and age than the L-band SAR data, from ALOS-2 PALSAR-2, highlighting an enhanced sensitivity of the long wavelength P-band signal with high biomass forests (>300 t/ha).

It is to expect that radiometric and polarimetric calibration as well as terrain normalization of the BIOMASS data will further improve the correlations with forest structural parameters. In addition, a significant improvement in sensitivity to high biomass forests is anticipated from the BIOMASS ground-notched data, which exclude ground contribution from the P-band signal. In summary, this study shows promising results from the early ESA BIOMASS L1B IOC amplitude data to advance global forest mapping from space and meet mission’s scientific requirements.

This work was conducted under ESA BIOMASS DISC as part of the BIOMASS In-Orbit Commissioning Programme.

Satellite imagery acquired in the last decade have substantially improved the knowledge of the spatial distribution of aboveground biomass (AGB) worldwide. Wall-to-wall maps of AGB impact downstream activities related to climate and vegetation modelling, national inventories and policy making. The start of operations of the BIOMASS satellite provides a novel data stream that in principle shall further improve the estimation of AGB from space. At the early stage of the mission, data from the Cal/Val phase are relevant to understand the contribution of BIOMASS to the quantification of AGB from space. Here, we intend to relate the Cal/Val datasets released by ESA to the set of predictors used by the Climate Change Initiative (CCI) Biomass retrieval algorithm, i.e., multi-temporal Sentinel-1 and ALOS-2 PALSAR-2 radar backscatter images and canopy height from the ICESat-2 LiDAR mission. Our scope is to assess similarities and differences between these data streams and understand the signatures the BIOMASS data. Ultimately, our goal is to assess how BIOMASS data can contribute to large-scale estimates of AGB within the context of the CCI Biomass framework. This work relies on the database of satellite observations gathered in CCI Biomass, which is updated regularly with the newest observations. A second objective of our study is to relate the BIOMASS data to CCI Biomass estimates of AGB. This is a purely explorative approach aimed at flagging potential caveats of the CCI Biomass dataset (e.g., due to lack of sensitivity of the C- and L-band predictors to biomass) and the BIOMASS data (e.g., due to ionosphere, geolocation accuracy, pre-processing etc.). This work is undertaken as part of ESA’s CCI Biomass project and EU’s NextGenCarbon project.

The ESA BIOMASS mission, successfully launched on 29 April 2025, is the first spaceborne P-band synthetic aperture radar (SAR) dedicated to continental-scale mapping of forest biomass, height, and disturbance. The first acquired datasets demonstrate excellent radiometric and interferometric quality, confirming the instrument’s unprecedented sensitivity to forest above-ground biomass density (AGBD) and vertical structure.

In this contribution, we present an intercomparison of AGBD maps over the Amazon forest, generated using an updated variant of the global BIOMASS AGBD mapping algorithm. As the primary AGBD predictors, we use three distinct datasets acquired during the commissioning phase:

(i) Level-1B backscatter coefficients,

(ii) Level-2A ground-cancelled backscatter, and

(iii) interferometric phase height at P-band.

As an independent reference, we employ X-band phase-centre height-based AGBD estimates, previously shown to exhibit a sensitivity to biomass comparable to that of airborne lidar.

We evaluate how the inclusion of Level-2A ground-cancellation processing and P-band interferometric phase height enhances the sensitivity of BIOMASS observables to AGBD and improves retrieval robustness relative to conventional backscatter-only approaches. Furthermore, we assess the accuracy and stability of biomass estimates derived directly from the P-band phase-centre height.

These preliminary results provide one of the first demonstrations of the BIOMASS mission’s capability for large-scale AGBD estimation, paving the way for robust, continental-scale biomass mapping and monitoring once the mission enters its operational phase in 2026.

This contribution describes the principles and implementation of a Level 3 (L3) product processor for ESA’s BIOMASS mission. This mission aims at reducing the uncertainty in the worldwide spatial distribution and dynamics of forest biomass, and will achieve this objective using a P-band SAR, providing global maps of forest biomass stocks, forest disturbance and growth [1].

In its dual-baseline interferometric operating phase, the BIOMASS mission will provide at each Global Cycle (GC), i.e. approximately every 7 months, a set of world-wide Level 2b (L2b) products consisting of maps of the Above Ground Biomass (AGB), Forest Height (FH) and Forest Disturbance (FD).

The L2b AGB and FH estimation processes being led independently, and at rather local spatial and temporal scales, it is very likely that output maps show some variability, related to the intrinsic uncertainty of L2b estimators, but also to potential exceptionally unfavorable factors, such as severe meteorological conditions (rain, wind), problematic propagation effects, or non-optimal baseline configuration.

The objective of L3 processing is to improve the consistency of L2b maps, by enforcing geophysical constraints through an iterative statistical regularization process. Three kinds of constraints are considered:

- spatial consistency is derived by comparing the spatial statistics of each product with autocorrelation function features computed over a wider neighborhood. Significantly different behaviors are to be penalized in order to guarantee spatially homogeneous estimates over undisturbed areas.

- temporal consistency is evaluated by observing estimates performed for different GCs, and by limiting the positive change rate (gain velocity) of AGB and FH parameters. One may note maximal gain rates are fixed according the considered geographical location and to the observed type of forest. Unlike gains, AGB and FH losses are not constrained, as they can happen in a very abrupt way.

- allometric consistency allows to mutually regularize AGB and FH fields using local, and forest-class specific relationships [2]. Allometric equations, as well as their associated dispersion, are directly estimated from L2b maps, at each GC, for each of the forest class provided by a land-cover map, and at the scale of an L2b tile, i.e. over regions of about 100 km x 100 km at the equator.

The regularization process is implemented under the form a Maximum Likelihood optimization, aiming to determine AGB and FH space-time maps which maximize a compound likelihood function, composed of losses terms related to L2b, spatial, temporal and allometric statistics [3].

A log-normal framework is adopted, which allows to represent this optimization as a very large, but sparse, system of linear equations.

At each new GC, the optimization process is run considering the freshly estimated L2b parameters in addition to the previously regularized fields. As a consequence, L3 estimate maps are expected to change significantly during the lifetime of the mission, with a quality level that increases with time.

The L3 processor delivers, at each GC and for each processed tile, L3 AGB and FH maps together with their level of confidence [4] and descriptors of their temporal evolution, auxiliary information related to the considered land-cover maps, and parameters of the BIOMASS allometric relationships at all dates and for all forest types.

An example of regularization, built from realistic AGB and FH maps estimated over Gabon, is provided and illustrates the capabilities of the processor to actually improve the consistency of L2b product maps.

[1] Quegan, S. et al. “The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space”, Remote Sensing of Environment, Volume 227, 2019, Pages 44-60, ISSN 0034-4257, https://doi.org/10.1016/j.rse.2019.03.032.

[2] Chave, J. et al. “Improved allometric models to estimate the aboveground biomass of tropical trees”, Global change biology, 2014, 20(10), pp.3177-3190.

[3] Tarantola, A. (2005) Inverse Problem Theory and Methods for Model Parameter Estimation. SIAM: Society for Industrial and Applied Mathematics, 342 p. https://doi.org/10.1137/1.9780898717921

[4] Keener, R.W., 2010. Theoretical statistics: Topics for a core course. Springer Science & Business Media.

.

.

The Product Algorithm Laboratory (PAL) for Biomass mission represents a key asset of the European Space Agency’s Multi-Mission Algorithm and Analysis Platform (ESA MAAP), a cloud-based platform that enables the processing and analysis of Earth observation data across multiple missions.

The PAL is built on CGI’s Insula platform, an advanced environment for Earth Observation (EO) data analytics that combines cutting-edge, production-ready technologies to deliver a flexible and interoperable cloud framework. This infrastructure underpins the ESA MAAP environment, currently supporting the Biomass and EarthCARE missions, while ensuring security, scalability, and efficient orchestration of processing services for a seamless and reliable user experience.

The Biomass MAAP enables scientists and developers to experiment with prototype algorithms, process large-scale mission data, and perform advanced analytical workflows in a shared, scalable, and reproducible infrastructure.

The PAL integrates direct access to the Biomass mission products, auxiliary datasets, and related validation data and provides scalable computing resources. It allows researchers to deploy and execute custom processing chains without managing underlying infrastructure. This capability fosters an iterative approach to algorithm refinement, from prototyping to pre-operational qualification.

Within this environment, researchers can perform comparative studies across multiple algorithm versions or missions, conduct sensitivity analyses to assess the impact of input parameters and assumptions, and perform advanced visualizations and analytics to better interpret data and algorithm behavior.

These capabilities promote knowledge exchange, algorithm harmonization and interoperability across missions, supporting the development of consistent and scientifically robust products within the ESA MAAP where tools, datasets, and users interact efficiently to empower EO data analysis.

By combining collaboration and experimentation, the PAL creates a framework that bridges scientific research and operational implementation, accelerating EO algorithm innovation and ensuring readiness for operational exploitation within ESA’s upcoming missions.

Importantly, the Biomass PAL is fully operational and open to researchers and developers, providing direct access to mission data and computing resources within a ready-to-use environment.

The X-band Synthetic Aperture Radar (SAR), with its short wavelength (~3 cm), offers exceptional spatial resolution and sensitivity to surface micro-structures—an attractive asset for precision agriculture. However, its limited penetration into vegetation and soil has historically constrained its utility for deeper biomass or subsurface retrievals.

Previous studies have demonstrated that X-band SAR data are particularly effective in distinguishing between surface roughness and tillage states over bare or sparsely vegetated soils (Baghdadi et al., 2002), owing to their enhanced sensitivity to micro-topographical variations. Nevertheless, several works have emphasized that this sensitivity decreases markedly in the presence of vegetation, as canopy volume scattering masks the soil contribution (Lopez-Sanchez & Ballester-Berman, 2009). More recently, multi-frequency comparative analyses have further confirmed that the X-band signal is predominantly sensitive to surface moisture and roughness, whereas deeper soil layers are better probed by lower-frequency systems (Zhou et al., 2025).

In this study, we explore two complementary aspects of repeat-pass PolInSAR data at X-band for agricultural applications: (1) the potential of repeat-pass interferometric phase coherence to reveal subsurface moisture and drainage features beneath cropped fields; and (2) the value of full polarimetry (orientation sensitivity, depolarisation metrics) to detect sowing ridges, tillage patterns, and crop growth stage via structural anisotropy of the soil–vegetation system.

Firstly, using TerraSAR-X Staring Spotlight acquisitions (single-polarisation, high resolution), we demonstrate that under specific seasonal and hydrological conditions, the interferometric phase (and coherence) signal exhibits sensitivity to subsurface moisture contrasts. In particular, we observe coherent phase-pattern anomalies aligned with known subsurface drainage networks beneath agricultural parcels—a surprising result, given the limited penetration of X-band. We interpret these signals as the result of dielectric contrasts in the shallow subsurface (wet vs dry soil), modulating the scattering phase behavior.

Secondly, employing an airborne, fully-polarimetric X-band sensor (SETHI), we investigate how polarimetric parameters (e.g., orientation angles and depolarization indices) vary with the presence or absence of sowing ridges and as a function of early crop growth stage. Our results show a clear correlation: fields with visible ridging manifest stronger anisotropy in orientation angle domains and lower depolarization, whereas mature crop covers produce more isotropic scattering and higher depolarization. Consequently, we find that the polarimetric signature can serve as a proxy for both soil preparation state (ridges vs flat) and early crop phenology.

Together, these findings illustrate the under‐exploited potential of X-band PolInSAR for precision agriculture: (i) as a sensor of subtle subsurface moisture heterogeneity via coherence/phase analytics, and (ii) as a structural probe of soil and crop architecture via polarimetry.

References

[1] Baghdadi, N., Holah, N., Dubois-Fernandez, P., Prévot, L., Hosford, S., Chanzy, A., ... & Zribi, M. (2004). Analysis of X-Band Polarimetric Sar Data for The Derivation of The Surface Roughness Over Bare Agricultural Fields. A A, 3, 2.

[2] Lopez-Sanchez, J. M., & Ballester-Berman, J. D. (2009). Potentials of polarimetric SAR interferometry for agriculture monitoring. Radio science, 44(02), 1-20.

[3] Zhou, X., Wang, J., Shan, B., He, Y., & Xing, M. (2025). Sensitivity of multi-frequency and multi-polarization SAR to soil moisture at different depths in agricultural regions. Journal of Hydrology, 133513.

This paper introduces Adaptive sum of Kronecker products for QUAntitative- Soil Moisture Anomalies retrieval (AKQUA-SMA), a novel polarimetric framework for SMA estimation. The core innovation lies firstly in the use of Temporal-SKP (T-SKP) [1], which exploits the temporal-polarimetric domain to isolate two scattering components: the moisture-related contribution, namely the Latent contribution, from the other scattering mechanism, namely Ground, i.e., above-ground scattering components. To do so, the SKP solution is chosen through exhaustive search as the one that minimizes the l1-norm of the error between the decomposed coherence values in the Latent structure matrix and the theoretical expected moisture related complex coherences. In particular, the search grid originates from the model proposed in [2], which has been slightly modified to account for a double-scattering mechanism. Then, the Ground component is chosen as the one that minimizes the phase residues, i.e., the most triangular scattering mechanism, and used for phase calibration similarly to what is done for the BIOMASS tomography. Finally, absolute N (zero-mean) SMA values are regressed by LS estimation from the estimated variations in the N(N-1)/2 InSAR pairs.

The AKQUA-SMA algorithm was applied to the Hydrosoil dataset [3], which was collected over a 20m × 58m agricultural field using a C-Band ground-based PolSAR (GB-PolSAR). The campaign aimed to simulate the frequent monitoring capability of the HydroTerra mission [3] for soil moisture and vegetation parameter retrieval. The data comprises two phases: the Barley Crop (March–June 2020), a Dual-Pol dataset (18,055 acquisitions), and the Corn Crop (July–November 2020), a Quad-Pol dataset (12,945 acquisitions), both acquired every 10 minutes. This SAR data is supplemented with essential ancillary information, including probe-based volumetric moisture, plant density, and crop height. Due to the limited field size, the entire area was treated as a single resolution cell. Processing utilized a full overlapping sliding window approach, scanning the dataset in steps of six acquisitions with five temporal samples overlapping between adjacent windows. The first results reveal good estimation accuracy, with a mode RMSE of 0.2% across the entire Corn dataset. Experiments involving the estimation of SMA through exhaustive search using the single polarization channels after phase linking are currently running, in order to assess the benefit of the multi-polarimetric approach w.r.t. the single-polarization one.

References

[1] Tebaldini, Stefano. ”Algebraic synthesis of forest scenarios from multibaseline PolInSAR data.” IEEE Transactions on Geoscience and Remote Sensing 47.12 (2009): 4132-4142.

[2] De Zan, Francesco, et al. ”A SAR interferometric model for soil moisture.” IEEE Trans actions on Geoscience and Remote Sensing 52.1 (2013): 418-425.

[3] Aguasca, Albert, et al. ”Hydrosoil, soil moisture and vegetation parameters retrieval with a C-band GB-SAR: Campaign implementation and first results.” 2021 IEEE Inter national Geoscience and Remote Sensing Symposium IGARSS. IEEE, 2021

Soil moisture is a crucial parameter in hydrology and agronomy applications [1,2] playing an important role in water resource and irrigation management. Synthetic Aperture Radar (SAR) data have been extensively used in several studies to retrieve the soil moisture across various land covers from bare to vegetated soil, including cropland, grassland, and forest [3,4]. Several scattering models have been developed to retrieve soil moisture beneath vegetation [5–7]. Using first–order radiative transfer (RT) backscattering models requires a complete description of vegetation structure. Thus, extensively used semi–physical models such as the Water Cloud Model (WCM) [8], rely on vegetation information (e.g., vegetation water content and biomass) derived from optical vegetation indices including LAI and NDVI. However, such indices provide a partial description of vegetation structure, leading to soil moisture retrieval errors. In this context, Deep Learning (DL) techniques can be synergically used with physics–based models to optimize their parameters and compensate the lack of vegetation structure description.

In this study, we apply the Physics-informed Residual Network (ResNet) model RTNet, proposed and validated using airborne data in our previous work [9], and specifically designed for accurate and physically meaningful soil moisture retrieval. Particularly, we extend the use of RTNet to spaceborne data from ALOS–2 acquisitions at HV polarization, together with multi-source remote sensing (RS) data, e.g., vegetation water content (VWC), soil texture, and weather information. While the HV polarization is used as input to the AI model, the HH polarization is employed, as described later, for the model optimization task.

The RTNet is designed to simultaneously perform two optimization tasks. The first one aims to estimate the parameters of a first–order RT model, namely the four scattering contributions (i.e., surface, double, volume, and triple scattering) and the attenuation term. The second task focuses on soil moisture retrieval starting from the optimized scattering components, inspired by the purely physical modeling approach proposed in [10]. Thus, two loss functions are defined: the first minimizes the difference between the measured total backscatter at HH polarization and the estimated one, computed as the sum of the optimized attenuated scattering components, while the second minimizes the difference between the retrieved soil moisture and in–situ measurements. These two losses are properly combined to guide the RTNet training, ensuring an accurate and consistent soil moisture retrieval by coupling it with the optimization of the scattering mechanisms based on physical modeling. To guarantee the physical validity of the estimation and to avoid ill–posed solutions, additional physical constraints are imposed within the loss function.

The methodology is validated using soil moisture measurements from the SCAN network dataset [11] and ALOS–2 L–band acquisitions aggregated to a 100 meter resolution over six test sites characterized by different land cover types. On the other hand, the optimized RT parameters, i.e., the scattering mechanisms, are evaluated through a comparison with well–established polarimetric decompositions, e.g., the Freeman–Durden three–component decomposition (FD3) [12], the Yamaguchi four–component decomposition with rotation of the coherency matrix (Y4R) [13], to assess the strengths and limitations of the proposed approach.

As a future development, the framework will be applied to L–band NISAR data, offering the opportunity to evaluate the robustness of the proposed framework against global soil-vegetation condition variabilities.

To conclude, this methodology goes beyond retrieving soil moisture. Instead, through the ingestion of data from heterogeneous sensors and the decomposition of the radar signal into its scattering components, it provides a comprehensive tool useful for investigating soil–vegetation interactions, seasonal changes, and other applications such as land cover characterization and crop type identification, opening new possibilities for development, training, and validation of explainable physics–informed AI models.

References

[1] Schaufler, G., Kitzler, B., Schindlbacher, A., Skiba, U., Sutton, M.A., & Zechmeister-Boltenstern, S. (2010). Greenhouse gas emissions from European soils under different land use: effects of soil moisture and temperature. European Journal of Soil Science, 61.

[2] Sheffield, J., Goteti, G., Wen, F., & Wood, E.F. (2004). A simulated soil moisture based drought analysis for the United States. Journal of Geophysical Research, 109.

[3] Hosseini, M., & Mcnairn, H. (2017). Using multi-polarization C- and L-band synthetic aperture radar to estimate biomass and soil moisture of wheat fields. Int. J. Appl. Earth Obs. Geoinformation, 58, 50-64.

[4] Baghdadi, N.N., El-Hajj, M., & Zribi, M. (2016). Coupling SAR C-Band and Optical Data for Soil Moisture and Leaf Area Index Retrieval Over Irrigated Grasslands. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 9, 1229-1243.

[5] Baghdadi, N.N., & Zribi, M. (2006). Evaluation of radar backscatter models IEM, OH and Dubois using experimental observations. International Journal of Remote Sensing, 27, 3831 - 3852.

[6] Fung, A.K., & Chen, K. (2004). An update on the IEM surface backscattering model. IEEE Geoscience and Remote Sensing Letters, 1, 75-77.

[7] Bracaglia, M., Ferrazzoli, P., & Guerriero, L. (1995). A fully polarimetric multiple scattering model for crops. Remote Sensing of Environment, 54, 170-179.

[8] Attema, E., & Ulaby, F.T. (1978). Vegetation modeled as a water cloud. Radio Science, 13, 357-364.

[9] Huang, X., Papale, L.G., Lavalle, M., Del Frate, F., Fattahi, H., Chan, S.K., Lohman, R.B., Xu, X., & Kim, Y. (2025). Optimizing the Radiative Transfer Model Using Deep Neural Networks for NISAR Soil Moisture Retrieval. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 18, 12697–12712.

[10] Hajnsek, I., Jagdhuber, T., Schon, H., & Papathanassiou, K. P. (2009). Potential of estimating soil moisture under vegetation cover by means of PolSAR.IEEE Transactions on Geoscience and Remote Sensing, 47(2), 442-454

[11] Schaefer, G.L., Cosh, M.H., & Jackson, T.J. (2007). The USDA Natural Resources Conservation Service Soil Climate Analysis Network (SCAN). Journal of Atmospheric and Oceanic Technology, 24, 2073-2077.

[12] Freeman, A., & Durden, S.L. (1998). A three-component scattering model for polarimetric SAR data. IEEE Trans. Geosci. Remote. Sens., 36, 963-973.

[13] Yamaguchi, Y., Sato, A., Sato, R., Yamada, H., & Boerner, W. (2010). Four-component scattering power decomposition with rotation of coherency matrix. 2010 IEEE International Geoscience and Remote Sensing Symposium, 1327-1330.

Observing changes in soil moisture with differential Synthetic Aperture Radar (SAR) interferometry (D-InSAR) has gained relevance due to shorter satellite revisit times, bigger swath widths, higher spatial resolution of SAR systems, and the introduction of L-band satellites. The geometric configuration of D-InSAR mitigates topographic effects, baseline-induced distortions, and geometric decorrelation.

In this context, the separation of volume, surface and dihedral scattering in the SAR signal is essential for limiting uncertainty in the accurate estimation of soil moisture. Using D-InSAR recent soil moisture forward models are able to account for varying soil states and surface and subsurface volume scattering combinations, although they assume the absence of topographical variations and vegetation between acquisitions [1], [2]. Whereas former can be approximately removed by using phase triplets, separating above ground volume from surface scattering remains a challenge. To address this, polarimetric information has been shown to be sensitive to vegetation [3].

To enhance soil moisture estimation we combine existing D-InSAR electromagnetic forward models with polarimetric information. Here, the copolar phase differences (CPD) [4] can been used to counteract the need of auxiliary geometries and fully-polarimetric systems. CPD has been shown to be highly sensitive to volume scattering due to the structural anisotropy of vegetation.

This analysis will focus on quantifying the performance of Zwieback et al. [2] at L-Band and HH/VV polarizations by deploying it on the recent AgriROSE-L dataset [5]. The limits of the model will be explored in respect to its parameterization, the degree of vegetation and numerous soil moisture levels. First, we test the model performance over bare ground. Secondly, we analyze the value of CPD for identifying vegetated and non-vegetated pixels. Third, we provide a first approach of integrating a CPD indicator into current soil moisture forward models.

References

[1] F. De Zan, A. Parizzi, P. Prats-Iraola, and P. López-Dekker, “A SAR interferometric model for soil moisture,” IEEE Transactions on Geoscience and Remote Sensing, 2013.

[2] S. Zwieback, S. Hensley, and I. Hajnsek, “A polarimetric first-order model of soil moisture effects on the DInSAR coherence,” Remote Sensing, vol. 7, no. 6, pp. 7571–7596, 2015.

[3] G. Anconitano, M. Lavalle, M. A. Acuña, and N. Pierdicca, “Sensitivity of polarimetric SAR decompositions to soil moisture and vegetation over three agricultural sites across a latitudinal gradient,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 3615–3634, 2023.

[4] V. Brancato and I. Hajnsek, “Analyzing the influence of wet biomass changes in polarimetric differential SAR interferometry at L-band,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 11, no. 5, pp. 1494–1508, 2018.

[5] H. I. Schauer, N. Basargin, and I. Hajnsek, “Airborne L-Band soil moisture retrieval over agricultural areas in preparation for ESA ROSE-L mission,” 2025.

Recently, the German Aerospace Center (DLR) conducted the AgriROSE-L 2025 (CROPEX 2025) airborne F-SAR campaign in preparation for the upcoming ESA ROSE-L mission. Multiple flights acquired a long and dense (6-day revisit) time series of fully polarimetric SAR acquisitions over agricultural areas between April and July 2025. Ground teams accompanied each flight to collect in situ measurements and record the soil and vegetation conditions. Agricultural areas are subject to rapid changes caused by crop growth, transitions of phenological stages, and alterations in soil conditions, including changes in soil moisture and roughness. The campaign provides a valuable dataset for analyzing the dynamics of the polarimetric signal over time and developing new methods for geophysical parameter retrieval.

In this study, we provide an early evaluation of the new dataset, focusing on soil moisture estimation, an essential climate variable (ECV) important for hydrology and agriculture. A significant challenge for SAR-based high-resolution soil moisture estimation is the presence of vegetation, which interferes with the ground signal. Here, the use of longer wavelengths (L-band) with deeper penetration is beneficial in minimizing the influence of vegetation. Additionally, using fully polarimetric data enables a certain degree of separation between the ground and vegetation contributions.

We apply a model-based tensor decomposition [1] to separate the signal into surface, dihedral, and volume contributions. The method jointly exploits polarimetric and temporal information, analyzes a time series of acquisitions, and provides estimates for different geophysical parameters, including soil moisture. We validate the accuracy across different crop types and growth stages using the in situ measurements acquired during the campaign.

References

[1] N. Basargin, A. Alonso-González, and I. Hajnsek, "Model-based tensor decompositions for geophysical parameter retrieval from multidimensional SAR data," Submitted to IEEE Transactions on Geoscience and Remote Sensing, under review.

The UAVSAR AfriSAR campaigns of 2016 and 2024 represent major milestones in the joint NASA–ESA–DLR effort to advance calibration and validation (Cal/Val) of the NISAR and BIOMASS missions, while laying the foundation for future radar and lidar missions such as NASA’s Surface Topography and Vegetation (STV). These international collaborations demonstrate how coordinated airborne radar and lidar acquisitions can strengthen cross-agency scientific and technical cooperation, fostering new applications of 3D remote sensing for ecosystem, geomorphological , and hydrological research.

The 2016 AfriSAR campaign in Gabon established a benchmark by acquiring extensive multi-baseline PolInSAR and TomoSAR datasets, complemented by airborne and spaceborne lidar. These data enabled groundbreaking studies—over 100 peer-reviewed publications—on tropical forest structure, biomass, and sub-canopy topography. The 2024 AfriSAR-2 campaign, extending coverage across Ghana, Cameroon, Gabon, Democratic Republic of Congo and the Republic of Congo, builds on this success to refine TomoSAR acquisitions and improving retrieval algorithms for L-band (NISAR) and P-band (BIOMASS) measurements across a wider diversity of landscapes. These datasets directly support the ESA BIOMASS Cal/Val project PP0104629.

To meet the growing analytical complexity of these campaigns, new tools such as Kapok (for multi-baseline PolInSAR and TomoSAR analysis) and CAPON (for adaptive tomographic focusing) are being developed to enhance vertical resolution and structural accuracy. These advances help bridge the algorithmic maturity gap between radar and lidar approaches and contribute directly to the scientific and technological objectives of STV, envisioned as a unified mission for global mapping of surface topography and vegetation structure.

Looking ahead, the upcoming TropiSAR 2026 campaign in Peru and Colombia will expand these Cal/Val activities to the Amazon and Chocó-Darien-central America region, providing cross-continental datasets for BIOMASS, NISAR and STV. This effort reinforces the growing NASA–ESA collaboration in algorithm development, data sharing, and applied science. The presentation will survey the datasets acquired during these campaigns, assess current processing and algorithmic capabilities, and outline future development needs—particularly in TomoSAR processing, canopy-ground separation, and hydrologic coupling—to fully realize the potential of radar missions for global forest and water resource monitoring.

Development and use of low frequency (VHF to UHF) imaging radars has increased in recent years, driven by the presence of flagship scientific programs at European level, such as the BIOMASS mission aiming to map forest height and above ground biomass globally, or by various scientific applications requiring solving FOPEN (Foliage Penetration) issues.

More particularly linked to the BIOMASS mission and to support calibration/validation activities in the tomography phase, a new TropiSAR airborne campaign will be conducted by ONERA on 2027. Objective will be to map and characterize the dense tropical forest cover and the underlying surface using tomoSAR mode at low frequency.

SETHI low-frequency SAR sensors are particularly well-adapted for such a mission where a high performance is required for SAR imaging, repeat-pass interferometric and tomographic measurements.

The proposed campaign will fly again over dense French Guyana tropical forest (Paracou, Nouragues and Rochambeau areas) with P-band, L-band and X-band SAR sensors. Possibility to fly simultaneously multi-band SAR sensors is also of main interest for this campaign: We can then compare multi-band results on same area, with same conditions (weather, vegetation state).

This new campaign will benefit our latest developments in softwares to exploit scientific data using PolinSAR and tomography technics to retrieve information on forest height, density and potentially ground topography.

We will expose in this contribution the campaign plan for TropiSAR-2 experiment, including schedule and tracks selection.

* Land vegetation is a large carbon store and represents opportunities to sequester additional carbon. While many Earth Observation missions aim to estimate forest biomass from space, their calibration and validation is critical. Ultimately trust in biomass maps requires accurate ground data. Supporting ground measurements and the people who make them is thus mission-critical for mapping and tracking Earth’s forest carbon. Building on decades of work from the global research community with a strong representation of partners from the Global South, the GEO-TREES initiative funds high quality ground data from a global network of reference sites, and to make these data openly accessible. 

* In this contribution, we report on the progress in community building, data acquisition, processing and delivery at over 40 biomass reference measurement sites. For each biomass reference measurement site, data acquired by the consortium partners includes plot inventory measurements at ≥10 hectares of forest, aerial laser scanning (ALS) coverage over ≥1000 ha of forests, and terrestrial laser scanning of the forest for ≥3 hectares.  

* We intend to provide the following data: (1) a 0.25-ha resolution aboveground biomass density (AGBD, Mg/ha) estimate for each tree inventory subplot, together with a variance estimate; (2) a 0.25-ha resolution canopy height (m) estimate for each tree inventory subplot, together with a variance estimate; (3) a 0.25-ha map of AGBD inferred from ALS and plot data, together with a pixelwise variance estimate; (4) a 0.25-ha map of canopy height inferred from ALS, together with a pixel-wise variance estimate; (5) a 0.25-ha resolution aboveground biomass density (AGBD, Mg/ha) estimate for subplot scanned with TLS, together with a variance estimate; (6) ancillary data for each site. We detail how plot-level and ALS data is processed to account for uncertainty and possible bias, based on open-access pipelines that are both reproducible and that can be used by the broader GEO-TREES community, using the ESA MAAP. When ready, the data will be accessible on the GEO-TREES data portal. 

* We emphasize the importance of involving research scientists associated with the sites in product validation plans. Not only do they provide essential high-quality data, they also offer invaluable insights about the peculiarities of the study sites which a mission validation plan would ignore at its peril. The establishment of GEO-TREES, a coordinated network of validation sites, is crucial for the success of biomass missions. In the future, it could also prove useful for the validation of other Earth observation missions aimed at quantifying forest-related geophysical measurements. 

Accurate estimation of forest carbon stocks is critical for monitoring ecosystem dynamics and assessing climate mitigation strategies. Terrestrial Laser Scanning (TLS) is a powerful tool for quantifying forest structure and aboveground biomass (AGB) with unprecedented detail. TLS point clouds can be used to produce individual-tree metrics (e.g. stem diameter, height) and quantitative structure models (QSM) to estimate total tree volumes. Local allometric models can then be parameterized to predict tree volume and biomass from diameter and height. However, there are still uncertainties in TLS-based biomass estimates, which propagate through successive analytical stages, ultimately affecting carbon stock estimates while scaling them up from tree to plot level. The accurate estimation of carbon stocks is a challenge due to the use of generic and regionally mismatched allometric models, which potentially have high uncertainty (~40%). The uncertainty arises from field measurement errors, structural metrics errors, and statistical model uncertainties. Therefore, the propagation of uncertainty on the biomass estimates travels at different levels.

GEO-TREES is an initiative to provide the calibration and validation data for the satellite biomass products. This study investigates the propagation of TLS measurement errors into carbon stock assessments across a subset of GEO-TREES reference plots (Barro Colorado Island (Panama), Pasoh (Malaysia), Amacayacu (Colombia), and Mpala (kenya)), representative of a range of tropical forest types.

We quantify sources of uncertainty arising from the TLS data acquisition and post-processing (Segmentation of individual trees and QSM-fitting) to the allometric model parametrization and evaluate their cumulative impact on AGB and carbon estimates. Point cloud processing, tree segmentation, and Quantitative Structure Model (QSM) reconstruction can further contribute to structural inaccuracies, particularly in complex or overlapping canopies. Subsequent derivation of individual tree metrics (e.g., stem diameter, height, branch volume) carries both measurement and model-fitting uncertainties. These propagate into the estimation of tree volume and biomass through the use of volumetric or allometric models, respectively. By employing error-propagation frameworks (Monte Carlo propagation), we identify the dominant contributors, such as the QSMs model fitting error (which can systematically overestimate small branches), to total uncertainty under varying forest conditions. The study further proposes a standardized protocol for quantifying uncertainty and collecting data for TLS-based biomass estimation within the GEO-TREES network. Our findings offer critical insights for harmonizing TLS methodologies and enhancing the reliability of ground-based carbon reference data across a wide range of tree sizes for the ESA BIOMASS mission.

Conventional field census measurements, such as diameter at breast height (DBH) or height, capture only limited aspects of three-dimensional distributions in forest structure. These measurements are often converted to aboveground biomass (AGB) estimates using allometric models. AGB estimates through allometric models are often considered as ground truth for the calibration and validation of spaceborne remote sensing products. These tree size-to-mass allometric models are mostly built on a selected sample of harvested biomass data, but are then often applied to trees that fall far outside the size or ecosystem range of the model calibration data. This can result in potential errors in downstream AGB products from satellite data.

Three-dimensional measurements from terrestrial laser scanning (TLS) have demonstrated that they can overcome the typical limitations of current allometric models and capture the spatial distribution of forest biomass. TLS measurements are increasingly becoming more routine, and GEO-TREES is an example of an initiative that builds on and complements existing long-term ecological plot networks by integrating TLS, airborne laser scanning, and forest inventory census to support the upscaling of aboveground biomass using satellite remote sensing.

Using 3D TLS data collected over a range of forest ecosystems, we illustrate the potential impact of the current issue of conventional allometric models. In our case study of Wytham Woods (UK), we demonstrated using TLS that its AGB is 1.77 times more than current allometric model estimates. We will present two solutions to this problem: (a) TLS can be used to estimate the volume of an individual tree and the entire stand in 3D directly. These volume estimates can be converted to AGB using wood density values. This approach also offers full traceability of the AGB of each tree; (b) TLS can be used to generate 3D tree models across the full size range of trees, which can then be used to create new allometric models that do not need to be extrapolated out of sample. We will further illustrate this solution by a recently constructed a new allometric model using TLS for Eucalyptus tereticornis, the dominant species at EucFACE, an ecosystem-scale mature forest free-air CO2 enrichment (FACE) experiment in Australia.

In both solutions (direct or indirect through new allometric models), TLS is essentially used to virtually harvest trees. Whereas the first approach can provide a deeper understanding of the AGB of all trees in a forest stand, it requires significantly more time to collect and process the data. The construction of new allometric models using TLS provides a practical way forward to improve estimates of AGB for calibration and validation of spaceborne AGB estimates using satellites such as ESA BIOMASS.

This work is dedicated to a starting Biomass Cal/Val project focused on temporal variability of P-band backscatter. In spite of an important literature on the subject, a comprehensive characterization of temporal variability remains challenging, especially from few observations (as for Biomass mission revisit) and when changes magnitude due to forest growth or loss can be easily confused with vegetation or soil moisture variations.

To address this question, the originality of our approach relies on the use of in-situ radar and flux data, respectively from the TropiScat-2 experiment and the Guyaflux tower located in the Paracou research area (French Guiana). The former consists in multifrequency (P, L and C bands) and quasi continuous (every 15 min) radar acquisitions from the top of the Guyaflux tower (ca 55m high), which enables to mimic satellite based acquisitions but with a much higher revisit.

The Guyaflux tower (labelled by ICOS as GF-Guy) has been generating meteorological and CO2, H2O and energy fluxes data since 2003, in order to determine the drivers of ecosystem CO2 source or sink strengths and evapotranspiration.

The TropiScat-2 instrumentation has been operating since 2018, making possible the development of data-driven and semi-empirical models of the radar backscatter variability with respect to a rather diverse range of meteorological conditions (especially for what concerns the intensity of dry and rainy periods in this tropical environment). These time-series models are mainly dedicated to backscattering intensity and temporal decorrelation, which can both serve to parameterize retrieval methods dedicated to the forest (dry) biomass and height estimation, contributing thereby to Biomass external calibration activities for this type of forest. Several examples of parameterization based on such time-series modeling will be presented, whether with fast and simple analytical formulations or with more complex microwave interaction models such as MIPERS-4D (a fully coherent model based on 3D forest representation).

For what concerns validation, the main objective is to compare the Net Primary Production (NPP) derived from the in-situ flux measurements with the forest biomass change estimates from the L3 products. Several observation periods will be tackled as the products delivery will follow, from which we should be able to gain in spatial resolution and better fit the Guyaflux footprint. This task will be also supported by the cross analysis between temporal variations of BIOMASS higher products (esp. L1) and TropiScat-2 radar data, together with the analysis of meteorological conditions and the closest in-situ forest biomass and height estimates. Given the first product delivery for French Guiana in 2027, our methodology will will be mainly illustrated with Sentinel-1, ALOS Scansar and simulated P-band data.

The ESA BIOMASS mission, recently launched as the first spaceborne P-band synthetic aperture radar (SAR), aims to generate global maps of forest aboveground biomass (AGB) and improve our understanding of forest contributions to the global carbon cycle. Despite its potential, the influence of canopy structural variability on P-band radar backscatter remains insufficiently understood, limiting the physical interpretability of BIOMASS observations.

In this study, we assess the sensitivity of P-band radar backscatter to forest canopy structure using the Michigan Microwave Canopy Scattering (MIMICS) model parameterised with terrestrial laser scanning (TLS) data. Our initial analysis focuses on four savanna woodland plots in Bicuar National Park, Angola, and one tropical rainforest plot in Lopé National Park, Gabon. Individual trees were extracted from the TLS data and reconstructed into quantitative structural models (QSMs) to derive detailed branch-level architectural parameters, which were then used to parameterise MIMICS simulations across all radar polarisations and incidence angles consistent with the BIOMASS observation geometry.

To investigate the relationship between canopy structure and P-band backscatter, we computed integrated metrics capturing fundamental physical properties relevant to electromagnetic interactions. Results indicate that tree size and structural complexity are primary factors of backscatter magnitude, with less structural complex generally producing stronger P-band backscatter. Simulated backscatter from the Gabon rainforest plot was further compared with BIOMASS Level-1 products over a nearby forested region to evaluate model performance and sensitivity.

This modelling framework enables us to explore and quantify the key structural parameters driving variability in the P-band radar response, and to examine potential variations across forest types and sensitivity to environmental factors such as soil moisture. Ultimately, the insights gained from this work will support the development of a physically-informed deep-learning approach aimed at assessing the accuracy and uncertainty of EO-derived biomass estimates from regional to global scales.

Recent studies show that forest degradation partly explain the decline in the forest carbon sink observed by top-down approaches. Tropical forest degradation is estimated to be responsible for 25% of forest carbon emissions, with approximately 20% of tropical forests disturbed by logging activities. In the Brazilian Amazon, CO2 emissions from fires and forest fragmentation reached 88% of gross deforestation emissions (Silva et al., 2021).

Improving knowledge of greenhouse gas emissions from these processes is essential to better understand and seize opportunities to mitigate climate change. However, many aspects of carbon loss associated with forest degradation remain insufficiently understood. Post-perturbation recovery, the impact of recurrent perturbations (especially understory fires and wildfires), and net biomass loss are still subject to active research.

In this paper, we assess the potential of the BIOMASS P-band radar sensor to evaluate the impact of forest degradation on tropical dense forests. In particular, We analyze the long-term impacts of degradation on forest biomass and structure as reflected in P-band backscatter, measuring post-perturbation resilience as a function of perturbation type, recurrence, and edge effects.

We use the extensive Deter system archive over the Brazilian Amazon as reference perturbation data. This dataset, produced by the National Institute for Space Research (INPE), comprises over 300,000 deforestation warnings and 180,000 degradation warnings spanning 2015-2025. This temporal coverage enables a chronosequence approach to build post-perturbation recovery curves for each perturbation type. By leveraging the Deter archive and chronosequence analysis, we aim to characterize recovery trajectories following various perturbations and assess the compounding effects of recurrent disturbances and edge effects on forest biomass. These results are expected to improve carbon emission estimates from forest degradation and enhance understanding of forest resilience in the Brazilian Amazon, with implications for REDD+ monitoring and tropical forest management.

Ongoing Earth observation missions bring unprecedented detail and comprehensiveness for understanding and quantifying the role of terrestrial ecosystem on the global carbon cycle. Yet, limited for longer term analysis given their contemporaneous shorter period in orbit. Legacy missions and datasets are essential to study dynamics and processes at longer time scales. Here, analysing global long-term datasets on vegetation aboveground biomass dynamics, spanning from 1992 to 2019, and atmospheric CO2 measurements, we study (1) the contribution of biomass dynamics to the atmospheric CO2 growth rate and (2) the CO2 fertilization effect on plant biomass.

Adopting a fully data-driven three-box model that simulates carbon dynamics within live vegetation, woody debris and soil organic carbon pools, and considers wildfires and spatio-temporal changes in primary productivity, we are able to explain over 60% of the observed variability in atmospheric CO2 growth rate over the period of 1997-2019 (R = 0.78, p-value < 0.05), with a low RMSE of 1.0 PgC yr-1. Our results show that, globally, lagged effects from heterotrophic pools account for 50% of the variability in atmospheric CO2 growth rate, exceeding three times the direct contribution of transient effects from the live biomass pool. These findings highlight the importance of quantifying tree mortality and cascading carbon release from litter and soils in shaping the terrestrial carbon balance.

We further leverage these Earth observations to isolate the specific contribution of elevated CO2 concentration to the biomass dynamics using both local multiple regression and residual methods. The approach is evaluated across an ensemble of dynamic global vegetation model simulations, showing low errors (RMSE: 0.04 and 0.02) and high correlation (R2: 0.79 and 0.88; p-value < 0.005). Globally, satellite-derived estimates indicate a global increase in AGB of 16.9% [13.9–18.8%] per 100 ppm rise in CO2 concentration. These observation-based estimates are close to those estimated by current land surface models (16.3 ± 5.0 %) but exceed estimates from global Earth system models (12.7 ± 6.5% for CMIP5, 13.2 ± 4.6% for CMIP6), suggesting an underestimation of Earth system models on the contribution of the land ecosystems in dampening anthropogenic CO2 emissions.

Overall, vegetation-atmosphere interactions from annual to decadal time scales show both the strong role of carbon loss processes and legacy dynamics, alongside a modest though larger CO2 fertilization effect on biomass when compared to global models. Ultimately, we contextualize these results on the possible future benefits from integrating BIOMASS, NISAR and the GEDI missions to better quantify and understand processes controlling growth, disturbance and recovery processes in terrestrial ecosystems.

An improved representation of the carbon and water cycle dynamics in terrestrial ecosystems underpins a large uncertainty reduction in modeling Earth system dynamics. The climate sensitivity of ecosystem processes controls land-atmosphere interactions and the overall atmospheric carbon uptake and release dynamics across scales. Local and Earth observations of vegetation dynamics are key for the evaluation of our understanding and support the quantification of process representation in model development. Previous research has shown the importance in undermining equifinality using multi-variate observation constraints, focusing water and carbon fluxes and stocks.

Long-wavelength radar backscatter provides unique insights into the dynamics of plant water and carbon dynamics when compared to optical EO products, as such, embeds the potential for constraining various parameters controlling local climate vegetation responses. In this study, we present an approach for assimilating Earth observation backscatter data in a terrestrial ecosystem model to improve estimates of vegetation parameters turnover rates. Among others, we focus on the information content of L-band ALOS PALSAR data in constraining vegetation dynamics at selected FLUXNET sites, where carbon and water fluxes and stocks are observed. Using a radar observation operator, a standard radiative transfer model, we design a model-data integration experiment to investigate the benefits of multiple backscatter observations versus unique above ground biomass to constrain model parameters.

The experimental setup focuses on the trade-off between information content from backscatter and uncertainties from observation operators versus sparse above ground biomass observations to constrain parameters controlling leaf and wood pool dynamics in vegetation. Current results indicate that the assimilation improves the estimation of aboveground biomass and constraints on turnover rates for both foliage and woody pools. Ultimately, data sparsity and availability exert control on model performance and prior model uncertainty on parameter constraints. Ultimately, this study highlights the potential of L-band backscatter to enhance vegetation carbon cycle modeling, emphasizes the added value of the upcoming ESA BIOMASS mission, and underscores the importance of integrating vegetation water dynamics into carbon models.