Conference Agenda

Accurate and spatially consistent estimation of forest above-ground biomass (AGB) is central to quantifying carbon stocks, monitoring ecosystem dynamics, and supporting climate mitigation strategies. While the BIOMASS mission will provide unprecedented structural information at P-band, its temporal and spatial sampling characteristics differ fundamentally from the high-revisit, global coverage delivered by Sentinel-1, NISAR, and the forthcoming ROSE-L mission. In preparation for the Copernicus SAR System-of-Systems, there is a pressing need to develop robust biomass retrieval frameworks tailored to the dense temporal sampling and near-zero baseline configurations of C- and L-band SAR missions. Within this study, we analyze Water Cloud Model (WCM) [1-3] and develop new strategy for biomass mapping. The proposed approach is designed to exploit both backscatter and repeat-pass interferometric observables in a physically consistent manner, enabling AGB estimation under operational C–L band configurations.

Beyond classical backscatter modelling, we introduce and emphasize an interferometric WCM-based strategy that leverages ground-to-volume scattering ratios derived from repeat-pass coherence observations. Under near-zero baseline conditions, interferometric coherence provides access to the relative balance between vegetation and ground scattering components. By linking interferometrically derived ground-to-volume ratios to their physically consistent WCM counterparts, biomass retrieval can be reformulated as a constrained estimation problem with a reduced parameter space. The multi-temporal framework improves parameter stability and mitigates noise and seasonal variability, while the integration of C- and L-band observations enables joint estimation of frequency-dependent attenuation behaviour under a shared biomass constraint. This integration of interferometric and backscatter information is expected to enhance sensitivity to biomass-related structural changes and extend the dynamic range beyond intensity-only approaches. Retrieval accuracy, dynamic range, and stability under varying environmental and acquisition conditions will be systematically analyzed. Particular attention is devoted to assessing the added value and the new methodological pathway enabled by the proposed interferometric WCM formulation. In addition, the possibility of integrating physics-guided learning approaches is explored. Rather than directly predicting biomass, neural networks are designed to estimate physically interpretable WCM parameters, which are subsequently used within the analytical inversion framework. This preserves physical consistency while enhancing robustness to noise, temporal decorrelation, and modelling simplifications. By unifying the proposed reduced parameter space WCM formulations within a multi-frequency, time-series context, this work establishes a scalable and operationally compatible pathway for C–L band biomass retrieval.

[1] J. Askne, P. Dammert, L. Ulander, and G. Smith, “C-band repeat-pass interfero-metric sar observations of the forest,” IEEE Transactions on Geoscience and Remote, Sensing, vol. 35, no. 1, pp. 25–35, 1997.

[2]. E. Attema and F. Ulaby, “Vegetation modeled as a water cloud,” Radio Science, vol. 13, no. 2, pp. 357–364, 1978.

[3]. M. Santoro, J. Askne, G. Smith, and J. E. Fransson, “Stem volume retrieval in boreal forests from ers-1/2 interferometry,” Remote Sensing of Environment, vol. 81, no. 1,pp. 19–35, 2002.

In the frame of a recently conducted study at DLR, the general applicability of a two-look ScanSAR technique for along-track deformation retrieval for the ROSE-L mission has been evaluated. In a next step, new further aspects of this approach are investigated.
The two-look ScanSAR approach consists in increasing the overlap between the bursts so that every point on ground is covered twice under different Doppler centroids. This is achieved in the conventional ScanSAR mode by reducing the burst time, which, consequently, leads to an azimuth resolution loss. With the ROSE-L system, however, two-look ScanSAR can be achieved by increasing the azimuth processed bandwidth of the bursts due to the larger Doppler bandwidth available. This is possible since the ROSE-L system uses multiple azimuth channels, hence having access to an effective azimuth bandwidth a factor five larger than the system PRF. Using a second pass over the same area, two interferograms of the same target area can be computed, referred to as two looks. The phase of each of the interferograms is directly related to the deformation in the corresponding line of sight. The along-track deformation is then retrieved from the phase of the differential interferogram of the two looks. The technique to compute a differential interferogram between interferograms obtained at different Dopplers is commonly known as spectral diversity or multi-aperture InSAR (MAI).
The main considered error contributions are thermal noise and azimuth ambiguities, but can also be extended to include ionospheric disturbances. It has also been shown in [1], that coherent azimuth ambiguities can introduce biases into the phase of an interferogram if certain conditions concerning power and coherence of the backscattered signals of the main target and of the ambiguity are met in a scene. The results of the two-look ScanSAR technique can thus also be affected by those biases. This contribution focuses on the mitigation of phase bias arising due to azimuth ambiguities in the ROSE-L two-look ScanSAR mode. To mitigate the biases, we first estimate the magnitude and phase of the coherent ambiguities in the interferogram using the first deformation estimate. Hereby, the effects of acquisition geometry and azimuth reconstruction algorithm have to be taken into account. In the next step, we coherently subtract the estimated ambiguities from the biased interferogram to mitigate the biases. The corrected interferogram bursts can then be used to obtain a more accurate estimate of the deformation on ground using the same two-look ScanSAR technique as in the first iteration.
To demonstrate the technique, the algorithm was added to the end-to-end chain implemented for the performance study. Hereby an artificial scene with known deformations is generated and two acquisitions matching certain coherence requirements are derived from this scene. In the next step, the characteristics of the ROSE-L system, such as PRF, antenna characteristics and ScanSAR acquisition mode are introduced into the simulated signals, in a way that coherent azimuth ambiguities are added to the main signal. Then, a post-processing chain is used to retrieve the along-track deformation. At this point, the deformation estimates are used to remove the ambiguity biases from the interferograms and to obtain a more accurate deformation estimate. Lastly, retrieved and inserted deformations are compared and the resulting measurement errors are evaluated.
The contribution will present the current status of the investigations with results obtained with the end-to-end simulator.
[1] M. Villano and G. Krieger, “Impact of azimuth ambiguities on interferometric performance,” IEEE Geosci. Remote Sens. Lett., vol. 9, no. 5, pp. 896–900, Sep. 2012, doi: 10.1109/lgrs.2012.2187271.

1. Introduction and principle

Synthetic Aperture Radar (SAR) interferometry is widely used since the 1990s to measure surface displacement [Massonnet et al., 1993] or topography. Due to the frequency variation of the chirp signal during the antenna emission of the radar pulses, high-resolution SAR images are acquired with a large bandwidth centered around a central frequency. The exploitation of this frequency variation, which can be retrieved by Fourier analysis and range spectrum processing, is called Multi-Chromatic Analysis (MCA). The MCA principle was first proposed in the 1990s [Madsen and Zebker, 1992]. However, first SAR satellites generations had rather narrow chirp bandwidths (about 20 MHz) which did not offer the possibility to apply this principle. Applications began with the availability of SAR satellites with larger bandwidths (larger than 100 MHz), in particular X-band satellites [Bovenga et al., 2011]. MCA can be applied to a single image in order to detect particular points whose phase variation is stable across frequencies. It can also be applied to SAR interferometry (InSAR): both images of an interferometric couple are sliced in different frequency sub-bands from which a set of interferograms are obtained. The interferometric phase variation across frequencies can be retrieved on frequency-stable pixels. Measuring this phase variation along the bandwidth virtually equates to obtaining an interferogram at a very low frequency (about the bandwidth frequency). The fringe rate is thus 50 to 100 times lower than the initial interferogram, which presents the great advantage of avoiding the noise-sensitive unwrapping step. This technique has been successfully applied to 3D reconstruction of highly contrasted topography areas [Bovenga et al., 2014; Libert, 2018].
Although MCA is a very powerful technique, its application remains quite limited for two reasons: results may be very noisy on most areas; moreover, sub-band filtering degrades the range resolution of the images by a factor proportional to the number of range sub-bands. In particular, degrading the resolution implies that each downsampled pixel merges the phase signal of several scatterers, complicating the retrieval of phase variation across frequencies. In this work, we thus limited the MCA principle to two sub-bands. Our results show that it offers a good compromise between frequency diversity exploitation and spatial resolution preservation.
However, the high noise level still remains an obstacle for exploiting this technique. Recently, several Deep Learning (DL) based approaches have been shown to effectively denoise SAR interferograms, in particular the self-supervised approaches MuChaPro [Denis et al., 2025] and its regularized version MuChaPro-R2 [Gaya et al., 2025]. In this abstract, we show the first demonstration of MuChaPro-R2 effectiveness to filter half-band interferograms and benefit from the MCA strategy to reduce phase ambiguities. This approach is illustrated on Cosmo-SkyMed images over Millau viaduct, in France.

2. Multi-Chromatic Analysis

Multi-Chromatic Analysis (MCA), also known as Split-Band Interferometry, is an extension of classical SAR interferometry where the spectrum of the pair of images is divided into multiple sub-bands along the range axis. Interferograms are then generated for each corresponding sub-band of the pair.
Because of the difference in “effective” central frequency between the sub-bands, the wavelength (and therefore the height ambiguity) varies. This leads to changes in fringe frequency. In the absence of noise, and for a given pixel, there is a linear variation of the interferometric phase among the sub-bands (barring the eventual phase wrapping) that is directly related to the height of the pixel.
In particular, in the case of two half-bands, this principle still holds. However, the estimation of the interferometric slope is less robust because of the reduced number of data points.
Nonetheless, half-band interferometry is quite effective in its own right. The bands are larger, leading to a reduced loss of range resolution and less merging of different phase contributions.
The two interferograms (the lower and upper half-bandwidth ones) can then be subtracted from each other. The resulting phase difference is equivalent to the phase of an interferogram generated with a larger wavelength and height ambiguity (by a factor of about 50 in our case), or equivalently with the same wavelength but a very small spatial baseline. So much so that the whole height variation on the image is typically contained within a single fringe, removing the need for phase unwrapping and all its associated issues.
In this study, we apply this technique to a pair of images acquired by Cosmo-SkyMed in Spotlight mode over the Millau viaduct in France. This very high structure (~300 m from top to bottom) is located in a contrasted topography area across the Tarn river gorges. The longest pile under the viaduct attains 245 m.

Fig. 1: Subset of the amplitude of the reference image over part of the Millau viaduct (slightly tilted vertical strip that crosses the center of the image). The Tarn River is the large dark horizontal strip at the top. The road is the thin dark line that runs parallel to the river, below the bridge. The bottom hill is covered by vegetation.
As shown in the results, the fringe frequency is greatly reduced in the difference of interferograms, leading to phase values being unequivocally associated with different heights (bridge deck, surrounding topography, road along the Tarn River). Thus, no unwrapping is required despite the important height variation in the area. It is possible to retrieve heights despite the presence of strongly decorrelated areas (due to vegetation and the river) that isolate high coherence areas and prevent phase unwrapping based on phase continuity.

Fig. 2: The two half-band interferograms. Note the small difference in fringe frequency between the lower-frequency half-band (left) and the higher frequency half-band (right). The whole area above the road is extremely noisy. Parts of the hillside are also highly decorrelated, mainly because of the vegetation and the 2 months temporal baseline.

Fig. 3: Difference between the half-band interferograms from Fig. 2. Note the absence of visible fringes. This is due to the greatly reduced fringe frequency, leading to the height variation in the whole image being represented within a single 2π interval. In particular, the bridge deck has a different phase value than the hillside and the road.

Yet, these promising results of MCA are somewhat tarnished by the noise present in the interferograms and carried on in the difference of interferograms. A filter capable of removing noise from the half-band interferograms should greatly improve performance, as shown in the following.

3. Interferogram denoising by self-supervised deep-learning: the MuChaPro-R2 algorithm

While traditional InSAR denoising methods based on statistical models and non-local filters allowed to better preserve structures, they are reaching their limits in terms of current requirements for accuracy and generalization. Deep learning offers new approaches by learning directly from the representations adapted to the specificities of the radar signal. Major methodological contributions were recently proposed to jointly denoise phase and coherence. Based on linear combinations of multi-channel data, these methods reduce the multi-channel restoration problem to a series of single-channel despeckling problems followed by an inversion, with an optional spatial regularization that stabilizes the estimation despite strong radiometric or temporal variations. These approaches have only been tested on classical interferograms. We propose here their first application to sub-band interferograms.

4. Proposed method: 2MuCh, or how MuChaPro filtering can benefit sub-band interferometry

The application of MuChaPro-R2 to each half-band interferogram (see Fig. 4) significantly reduces the noise.

Fig. 4: Half-band interferograms filtered using MuChaPro. The small difference in fringe frequency is preserved. The top of the image is still noisy. The hillside and the bridge deck have greatly reduced noise, although some noisy areas survived.

As a result, the difference of interferograms (see Fig. 5) is also denoised. Separate homogeneous surfaces are more easily identifiable (hill and bridge), as is their relative height.
While low coherence regions still isolate high coherence areas (for example the red spot in the upper left corner of Fig. 6 right), the usable area is greatly increased.

Fig. 5: Difference between the filtered half-band interferograms from Fig. 4. The phase signal is clearly enhanced compared to Fig. 3. A slight phase gradient is now visible from the top of the hill to the road near the river.

Fig. 6: Difference of interferograms over a larger area centered on the viaduct. The noisy difference is on the left, the filtered one is on the right.

5. Bibliography

Massonnet, D., Rossi, M., Carmona, C., Adragna, F., Peltzer, G., Feigl, K. & Rabaute, T. (1993), ‘The displacement field of the Landers earthquake mapped by radar interferometry’, Nature 364, 138–142

S. N. Madsen and H. A. Zebker, "Automated Absolute Phase Retrieval in Across-Track Interferometry," [Proceedings] IGARSS '92 International Geoscience and Remote Sensing Symposium, Houston, TX, USA, 1992, pp. 1582-1584, doi: 10.1109/IGARSS.1992.578639.

F. Bovenga, V. M. Giacovazzo, A. Refice, D.O. Nitti, N. Veneziani “Interferometric Multi-Chromatic Analysis of High Resolution X-Band Data”, Fringe Workshop, Frascati, 2011.

Bovenga, F., Rana, F. M., Refice, A. & Veneziani, N. (2014a), “Multichromatic analysis of satellite wideband SAR data”, IEEE Geoscience and Remote Sensing Letters 11(10), 1767– 1771.

Libert, L. (2018). Towards operational use of combined Split-Band Interferometry and Multidimensional Small Baseline Subset: application to geohazard monitoring in the Kivu region [Doctoral thesis, ULiège - Université de Liège].

L. Denis, E. Dalsasso and F. Tupin, "Just Project! Multichannel Despeckling, the Easy Way," in IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1-11, 2025, Art no. 5204311

V. Gaya, L. Denis, B. Pinel-Puysségur and F. Tupin, “Self-supervised interferogram restoration by regularized inversion of despeckled projections”, submitted to IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) techniques consist in leveraging the temporal information provided by the time-series to improve the quality of each interferogram of the series. Leveraging the coherency allows for the measurement of Earth deformation up to millimeters accuracy. In this context, Interferometric Phase Linking (IPL) is a technique used to denoise the phase of SAR images by leveraging all possible pairs of interferograms within the time series. Existing methods from the state of the art have mostly focused on maximum-likelihood and least-squares fitting formulations. We reformulated the task as a covariance matrix fitting problem (COFI), where the aim is to recover the expected InSAR phase structure from a noisy estimate of the covariance matrix of a pixel patch. Such framework leaves open a choice regarding the matrix distance that will define the notion of optimal fitting. Previous work of ours analysed the use of the most classic Euclidean and Kullback-Leibler distances [6]. In our most recent work we have continued our study by exploring the use of several Riemannian distances on the space of covariance matrices: Affine Invariant, Log-Euclidean, and Bures-Wasserstein distances, and have derived an optimization algorithm to solve the corresponding fitting problems. After having presented our simulation results in [3], illustrating the interest of these distances in terms of estimation accuracy and computational complexity, we have applied the approach to areas of interest on Sentinel-1 and TerraSARX imagery. Because of the intrinsic characteristics of SAR imagery, statistical spatial homogeneity is typically assumed within small neighborhoods. This has naturally led to our next step: an IPL optimisation that integrates a spatial regularisation step through an Alternating Direction Method of Multipliers (ADMM) framework. We report the results obtained with the COFI methodology using the newly studied distances, alongside our first findings on the spatial regularisation step of the ADMM framework.

The framework

From a given datacube of p co-registered SAR images, we consider a local multivariate pixel patch {x_i}ⁿ_i=1, with x_i ∈ C_p, ∀i ∈ [[1, n]]. Each sample x_i contains the complex-valued time-series (in chronological order) of one pixel over the p snapshots. We assume that the patch is homogeneous, i.e., that the set {x_i}ⁿ_i=1 contains n pixels with similar scattering and statistical properties.

Interferometric phase linking (IPL) consists in estimating the complex phase vector w from the sample set {x_i}ⁿ_i=1 [4].The covariance fitting phase linking (COFI-PL) approach [6] involves the construction of a Covariance Matrix from the sample set and fitting it to any given plug-in estimate of the covariance matrix.

In our initial work, the focus resided on the choice of the matrix distance. Complex-valued covariance matrices belong to the space of Hermitian positive definite matrices. Endowing this space with a metric yields a Riemannian geometry for covariance matrices. In particular, the unit-modulus constraint on the complex phase vector implies that it belongs to a torus, and the optimisation was thus performed on this manifold. A full description and the corresponding simulation results for these optimisations can be found in [3].

Due to the intrinsic characteristics of SAR imagery, statistical homogeneity is typically assumed within small spatial neighborhoods. This has motivated the next step: combining the IPL cost function with a spatial regularization term. However, the IPL problem involves the construction of a Covariance Matrix estimate that yields the temporal correlation between the phases, and thus, is not a simple and explicit function of the samples. This prevents the use of spatial regularization penalty terms, as it would create an intricate interdependence between solutions of each subproblem, resulting in unscalable algorithms. The issue is tackled using the Alternating Direction Method of Multipliers (ADMM) framework [2]. The spatial regularisation used is a 2-dimensional Total Variation (TV) methodology [1]. We’ll be presenting initial results related to the TV step of our ADMM problem.

The Dataset

The study areas analysed are the area of Kahauale’a Natural Reserve in the island of Hawai’i using a time-series of 10 images acquired between the 23rd of November 2024 and the 28th of April 2025, and the area of Mexico City using a time-series of 15 images acquired between the 11th of March 2019 and the 21st July 2019. The displacement maps achieved following the COFI optimisation step were compared with displacement data provided by 17 GPS stations for Hawaii and 5 GPS stations for Mexico City monitoring the area on ground, provided by the Nevada Geodetic Laboratory GPS Networks Map [5]. The SAR data used in the analysis were acquired from C-band Sentinel-1 and X-band TerraSAR-X products.

Initial Findings on Real Data: GPS Vs. COFI InSAR Line-Of-Sight Displacement

The main focus of improving image quality is to provide a better estimate of the earth displacement. An analysis was performed that compared the displacement estimates of the sensor before and after optimisation with that of the ground data [5]. The displacement shown in the InSAR displacement map does not represent the true ground motion, but rather the displacement component observed by the radar, i.e. the projection of the ground displacement onto the radar line of sight (LOS). Therefore, the GPS displacement results had to be projected on the radar’s LOS in order for the comparison to be consistent. Boxplots summarizing the LOS displacement errors, aggregated for each GPS station analysed over all acquisition dates for the study area of Mexico City were analysed. The application of the COFI optimisation to the interferometric map leads to a substantial improvement in estimation performance relative to the original interferometric results. Moreover, the Euclidean-Log (LE) distance provides on average lower error rates and hence a higher accuracy of the displacement estimation compared to the same COFI method using the Kullback-Leibler (KL) distance.

Although the new tested distances provide an improvement, this improvement is very slight. Our current work consists in combining the COFI IPL with a Total Variation type (TV) spatial regularisation, through an ADMM. We will be presenting results not of the entire ADMM framework but of the TV spatial regularisation step of the ADMM.

References

[1] Alvaro Barbero and Suvrit Sra. Modular proximal optimization for multidimensional total-variation regularization. J. Mach. Learn. Res., 19(1):2232–2313, January 2018.
[2] Stephen Boyd, Neal Parikh, Eric Chu, Borja Peleato, and Jonathan Eckstein. Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends in Machine Learning, 3:1–122, 01 2011.
[3] Elena Grosso, Florent Bouchard, Frédéric Brigui, Guillaume Ginolhac, and Arnaud Breloy. Interferometric phase linking with riemannian distances on covariance matrices. In Proceedings of the 33rd European Signal Processing Conference (EUSIPCO), pages 2602–2606, 09 2025.
[4] Dinh Ho Tong Minh and Stefano Tebaldini. Interferometric phase linking: Algorithm, application, and perspective. IEEE Geoscience and Remote Sensing Magazine, 11:46–62, 09 2023.
[5] Nevada Geodetic Laboratory. GPS Networks Map, University of Nevada, Reno. 2025.
http://geodesy.unr.edu/NGLStationPages/gpsnetmap/GPSNetMap.html.
[6] Phan Viet Hoa Vu, Arnaud Breloy, Frédéric Brigui, Yajing Yan, and Guillaume Ginolhac. Covariance fitting interferometric phase linking: Modular framework and optimization algorithms.

Presently Interferometric Synthetic Aperture Radar (InSAR) methodologies typically rely on the processing of radar images acquired by single satellite missions, characterized by fixed temporal resolutions and specific spatial coverage. This reliance on isolated data sources poses significant challenges for civil engineering firms and public authorities who require high-precision, multi-dimensional data to assess structural stability and detect potential geohazards. Standard InSAR techniques are often hampered by inadequate revisit frequencies, restricted data diversity, and spatial resolution gaps, all of which can lead to delayed responses to critical infrastructure issues.

In the framework of the ESA-InCubed project, Cross Band Insights (CBI), an advanced InSAR monitoring solution designed to merge multi-mission SAR data has been developed. This hybrid approach leverages the complementary strengths of diverse satellite missions. By integrating these sources, CBI overcomes the sensor-specific limitations that often hinder effective ground deformation monitoring, providing a comprehensive view that encompasses Line-of-Sight (LOS), vertical, and horizontal displacement components.

One of the core aspects of the CBI project is to move the data-merging operation further back into the processing chain. Unlike current commercial products that process different satellite missions independently and only fuse results at the final visualization stage, the CBI workflow performs the critical final processing steps jointly.

A primary technical objective of the project is the systematic refinement of individual LoS products. This is achieved by shifting data integration into the heart of the processing workflow through joint Atmospheric Phase Screen (APS) estimation. Traditional methods estimate the atmospheric noise for each satellite track separately; however, CBI combines time series with different lengths or varying noise statistics to execute a motion decomposition "on the fly”.

Residuals from this joint model are utilized to filter the APS with significantly higher precision than single-geometry approaches. This joint estimation leads to more accurate and consistent time series by maximizing the spatio-temporal coverage and plugging refined data back into the original LoS products to mitigate systematic errors. For the end-user, this means that even the primary LoS measurements are "cleaned" by the presence of other satellite data, resulting in a more reliable baseline for structural health monitoring, preserving spatial and temporal resolution of the sensor used.

CBI defines also a generalized approach to motion decomposition, translating 1D LoS measurements into actionable Vertical (Up-Down) and Horizontal (East-West) displacement components. Furthermore, the system is designed to support the estimation of the North-South component if sufficient geometric diversity—such as the availability of ascending, descending, and left-looking or Mid-Inclination Orbit (MIO) data—is present.

This generalized model exploits mission-specific strengths, such as utilizing the broad coverage of C-band as a stable regional baseline while integrating the high-density sensitivity of X-band for complex structural monitoring over localized assets. The integration accounts for the differing accuracies and wavelengths of varied SAR sensors, ensuring proper weighting according to their native stochastic properties.

To support professional data interpretation and risk assessment, CBI implements a comprehensive error propagation model. This model tracks measurement accuracy through every processing step—from the initial spatial re-sampling on a common geo-grid to the final geometrical decomposition. By characterizing both random noise and systematic error sources, the system generates precise accuracy estimates (Error Bars) for every data point in the final time series.

To further assess the processing concept a verification activity is planned. The workflow will be operated using controlled inputs (deformation patterns, atmospheric phase screens, level of noise) and the results will be analyzed evaluating the robustness of the analytical model behind the algorithm and identifying constraints related to multi-band input. Then the workflow is operated using both simulated data and real-world case studies to evaluate the robustness of the underlying analytical models.

The approach has been finally tested on a real data scenario that includes Sentinel-1, Cosmo-SkyMed and TerraSAR-X datasets. The results have shown the possibility of enhancing the precision of the APS mitigation for the datasets having short temporal coverage. Moreover, generalized motion decomposition merging the data of all three different missions has been performed.

Cross Band Insights has as a goal to move InSAR technology from mission-dependent results to a source-independent, high-fidelity monitoring service. This is supposed to be achieved by improving LoS quality, generalizing motion decomposition, and providing rigorous error characterization. The solution offers the final users precise insights needed for sustainable and safe infrastructure management.