Conference Agenda

Spaceborne SAR interferometry has traditionally relied on large satellites operating in carefully designed near-frozen repeating orbits to ensure stable interferometric baselines and predictable geometric evolution. Achieving comparable interferometric stability with agile microsatellite platforms presents additional challenges. Smaller satellites experience stronger sensitivity to atmospheric drag relative to mass, more rapid orbital element evolution, and tighter propulsion resource constraints, leading to increased baseline drift if not actively controlled. These effects, combined with the absence of classical frozen-orbit tuning, can result in greater orbital variability and more rapidly changing interferometric geometry.

In this contribution, we present recent advances in operational interferometry using the ICEYE X-band microsatellite constellation, highlighting the transition from single repeat-track satellites with limited geographic coverage and wide orbital tubes to a proliferated Daily Ground Track Repeat (DGTR) architecture with significantly improved baseline stability and expanded global coverage.

Earlier ICEYE repeat-track satellites enabled daily coherent acquisitions over selected regions; however, total normal baseline variation over extended periods could reach tens of kilometers, limiting systematic time-series analysis. The current constellation configuration includes four satellites operating in independent 24-hour repeating ground track orbits. These satellites operate in independent 24-hour repeating ground track orbits and are phased to expand aggregate geographic coverage while minimizing redundancy. The result is nearly two-thirds of global land areas being accessible with daily revisit at the equator, and higher revisit density toward polar latitudes. This architecture represents a substantial increase in coherent coverage compared to earlier isolated repeat-track implementations.

Although these orbits are not yet fully frozen, active orbit control strategies constrain baseline evolution within a significantly narrower envelope than previous generations. During uninterrupted monthly acquisition campaigns, total normal baseline variation remains below approximately 850 m. Consecutive daily acquisitions exhibit perpendicular baselines well below 200 m and typically substantially smaller. This controlled baseline regime enables predictable interferometric geometry over extended time periods and supports systematic time-series processing despite operating outside classical frozen-orbit configurations. Ongoing orbital optimization efforts continue to further reduce baseline variability.

The proliferated DGTR architecture enables rapid stack accumulation. With daily revisit, interferometric stacks exceeding 100 acquisitions can be formed in slightly more than three months, significantly accelerating the time required to reach robust Persistent Scatterer Interferometry (PSI) solutions. High temporal sampling density provides several advantages: improved separation of atmospheric phase screen components, enhanced temporal coherence estimation, reduced phase aliasing risk for rapidly deforming targets, and increased sensitivity to transient or accelerating deformation phenomena. This sampling density is particularly valuable in scenarios approaching critical failure conditions, such as infrastructure instability, mining-induced subsidence, or landslide acceleration, where deformation rates may evolve on timescales shorter than conventional satellite repeat intervals.

In addition to repeat-track deformation monitoring, the constellation supports high-resolution spotlight imaging modes with ground resolutions down to approximately 25 cm. These coherent acquisitions enable fine-scale Coherent Change Detection (CCD), supporting detection of subtle anthropogenic or environmental changes that may not be discernible in wider-swath products. The combination of high spatial resolution and daily revisit enhances monitoring capability for rapidly evolving ground conditions.

Representative results are presented including:

• Differential InSAR time-series over active volcanic regions

• Persistent Scatterer deformation monitoring in urban and mining environments

• High-resolution coherent change detection examples demonstrating fine-scale surface modifications

To facilitate independent validation and algorithm development within the InSAR community, selected interferometric stacks from DGTR satellites are being released under a CC BY 4.0 open license as part of ICEYE’s Open Data initiative. By providing sustained daily-repeat stacks under controlled but non-frozen orbital conditions, this initiative supports benchmarking of time-series methodologies and encourages broader scientific engagement with microsatellite-based interferometry.

These results demonstrate the maturation of agile microsatellite SAR from early repeat-track experimentation toward a managed, proliferated daily repeat architecture capable of delivering sustained, high-quality deformation and change-monitoring products at regional to near-global scale.

Synspective has launched and operated seven small Synthetic Aperture Radar (SAR) satellites, "StriX," since the launch of its first satellite in 2020, as of the end of 2025. The company is proceeding with the manufacturing and development of its satellites to construct a constellation of over 30 satellites starting in 2028, which will enable the observation of any point on Earth and the delivery of analysis-ready data within approximately one hour of acquisition.

Key features of the small SAR satellite constellation are high resolution and high frequency. The best spatial resolution currently available from StriX is approximately 0.5m in ground range and 0.25m in azimuth. The shortest revisit time is one day, which enables daily ground displacement measurement, a capability impossible with conventional SAR satellites. StriX satellites utilize two types of orbits: Sun-Synchronous Orbit (SSO) and Mid-Inclination Orbit (MIO). While MIO precludes observation of high-latitude regions, it increases the observation frequency in mid-latitude areas. Furthermore, MIO offers sensitivity to the North-South component of displacement, which is not available with SSO InSAR, making it possible to reconstruct the 3D displacement field.

Although StriX currently does not maintain strict orbit control and cannot continuously acquire data applicable to Interferometric SAR (InSAR), there are specific periods when the perpendicular baseline shortens. During these opportunities, we successfully acquired datasets suitable for continuous displacement measurement and InSAR-based topographic measurement. We introduce three case studies below.

1. Validation of mm-Level Measurement Accuracy using Time-Series InSAR Analysis with Consecutive 1-Day Repeat-Pass Data

From December 13 to 23, 2025, when the baseline of StriX-4 (MIO) was short, we acquired consecutive interferometric data to demonstrate and validate the accuracy of daily InSAR time-series analysis. The observation conditions were Ascending, Left-looking, Incidence Angle 39°, and Azimuth Angle 69° (East-Northeast). The maximum perpendicular baseline was 1.2 km, which is sufficiently smaller than the critical baseline of 32 km. The observation mode was Staring Spotlight 4 (600 MHz bandwidth), with a spatial resolution of approximately 0.5 m in ground range and 0.25 m in azimuth. The Area of Interest (AOI) was Tsukuba City, Ibaraki Prefecture, Japan, where five Corner Reflectors (CRs) were installed. The height of two CRs was artificially changed by 1–2 mm seven times in total. CR displacement was measured by levelling. The maximum cumulative height change was approximately 1 cm.

The results of the InSAR time-series analysis and levelling showed good agreement with the artificial displacement, with the difference in displacement converging within a standard deviation of 1 mm at all points. This confirmed that daily InSAR with StriX can measure displacement with 1 mm accuracy at a high frequency of one-day intervals.

2. High-Resolution Pixel Offset Time-Series Analysis

The Pixel Offset (PO) method uses two SAR intensity images acquired under the same observation geometry to measure ground displacement that occurred between the two epochs through detailed image coregistration. Unlike InSAR, it does not require phase unwrapping, making it robust for measuring displacement with large spatial gradients, and its effectiveness has been demonstrated in capturing m-level displacement associated with large earthquakes. The achievable measurement accuracy is proportional to the spatial resolution of the SAR image, generally estimated to be about 1/10 of the resolution. Since conventional high-resolution SAR data typically had a spatial resolution of about 3 m, the measurement accuracy achieved by the PO method was limited to a few tens of centimetres, inferior to the mm-cm order accuracy achieved by InSAR. Furthermore, the need for a certain window size for detailed coregistration based on cross-correlation resulted in lower spatial resolution compared to InSAR, with measurable displacement scales generally considered to be 100 m or more. However, small SAR satellites are making higher-resolution data than conventional systems abundantly available. It is expected that the use of high-resolution data will enable cm-level displacement measurement even with the PO method.

We analyzed the Port of Karachi in Pakistan. The port is equipped with floating-roof tanks whose roofs move up and down depending on the contents' volume. The tanks have a diameter of approximately 25–50 m. StriX-2 (SSO) acquired 10 days of consecutive data from February 13 to 22, 2025. The observation conditions were Descending, Left-looking, Incidence Angle 32°, and Sliding Spotlight 1 (1 m resolution). The maximum perpendicular baseline was about 7 km, and the critical baseline was about 13 km.

In conventional PO analysis for wide-area ground displacement measurement, relatively large window sizes, such as 32x32, are often used to reduce noise. However, for targets with strong scattering intensity and characteristic structures, as in this case, a sufficiently high correlation was obtained even with a small window size of 8x8. By setting the sampling interval to half the window size, we calculated a high-resolution offset field of approximately 3.2 m. Even for the pair with a perpendicular baseline of about 7 km, a high correlation coefficient was obtained around the tanks, and clear displacement was captured. The theoretical standard error of the measured value depends on the correlation coefficient, but it was estimated to be smaller than 1/10 of the spatial resolution, at several centimetres, around the tanks.

Since redundant displacement values from multiple pairs are obtained from all combinations of consecutive observation data, time-series analysis can be applied, similar to unwrapped InSAR images. The time-series analysis results yielded displacement time series believed to reflect the vertical movement of the roof for specific tanks. Although ground truth data was not available, the estimated error from the consistency between redundant multiple pairs in the time-series analysis was generally less than 10 cm, and around 1–2 cm in good areas.

This result demonstrates the potential of PO analysis using high-resolution SAR data to estimate structural displacement with spatial resolution and measurement accuracy comparable to InSAR.

3. High-Frequency Monitoring and Topographic Measurement of Changes Associated with Volcanic Eruption Activity

On June 22, 2025, Mt. Shinmoedake, located in the Kirishima mountain range on the border of Kagoshima and Miyazaki prefectures in Japan, erupted for the first time in seven years. The initial eruption occurred on the northeast side of the crater, followed by continued active eruptive activity. Eruptions were also confirmed on the southeast side of the crater around July 3, and the plume reached 5,500 m above the crater rim on August 28. The series of eruptions continued until September 8.

The rapid assessment of the crater's location and shape is crucial for estimating the distribution direction of volcanic ejecta, the flow direction of pyroclastic flows, and the evaluation and prediction of eruptive activity. Traditionally, visual observation and optical images from helicopters or aircraft have been frequently used, but there are challenges: flight is impossible in bad weather, and the crater cannot be fully assessed at night or when the plume is active. SAR images, on the other hand, have the advantage of observing surface conditions regardless of weather or plume effects.

For this eruptive activity, observations were conducted by ALOS-2/4, acquiring a total of 13 data sets between June 28 and September 25, 2025, capturing changes around the crater. However, SAR satellite observation opportunities are limited to when the satellite passes over the target area, making agile observation difficult. ALOS-2/4 is limited to two observation times per day, around 0:00 and 12:00, and the revisit time is 14 days, making continuous daily observation impossible. Furthermore, the maximum spatial resolution is about 3 m, which limits the detection of local topographic changes smaller than 10 m.

For this eruption, Synspective conducted high-frequency, high-resolution observations using three StriX satellites (StriX-2, 3, and 4), capturing detailed surface changes associated with the eruptive activity. A total of 63 data sets were acquired between June 28 and October 1, 2025, with an average observation interval of approximately 1.5 days. The spatial resolution ranged from a lower resolution of 1 m to a higher resolution of 0.25 m, capturing local changes that were difficult to detect with ALOS-2/4 at high spatio-temporal resolution.

StriX-2 and StriX-3 employ SSO, similar to conventional SAR satellites like ALOS-2/4, resulting mainly in observations from the East-West direction. StriX-4, however, employs MIO, enabling observations including the North-South component. The combination of ascending/descending and left/right-looking observations allows for observations from up to eight directions. Integrating data from multiple directions can mitigate the effects of layover and shadow in areas with steep topography.

During the aforementioned observation period, data with baseline conditions suitable for InSAR were almost unavailable. However, in December 2025, after the eruptive activity subsided, StriX-4 reached orbital conditions suitable for InSAR, and one-day interval interferometric data of Mt. Shinmoedake were acquired. This data is applicable for DEM generation.

The InSAR phase includes contributions from atmosphere, ground displacement, topography, and errors due to decorrelation. Single-pass InSAR using two simultaneous satellites does not suffer from atmospheric noise, ground displacement, or temporal decorrelation, thus containing only the topographic component, which allows for high-precision DEM estimation. Conversely, with repeat-pass InSAR, which has a time difference, components other than topography generally become sources of error, making high-precision DEM estimation difficult.

However, under limited conditions, it is possible to extract high-precision topographic information from repeat-pass InSAR. First, to minimize the effects of ground displacement and temporal decorrelation, pairs with a sufficiently short time interval are selected. StriX's one-day interval pairs are suitable in this respect. Next, an appropriate perpendicular baseline pair is selected, which has sufficient sensitivity to topography and sufficiently small geometrical decorrelation effects. Since sensitivity to topography is inversely proportional to the wavelength, the X-band tends to have higher sensitivity than the C-band or L-band. A perpendicular baseline of about 500 m is optimal in this case, as the height ambiguity is about 15 m, providing sensitivity to heights of a few meters. Furthermore, to separate atmospheric error and the topographic component, we focus on the difference from an existing DEM. If the height change at the observation date from the existing DEM is local, and since atmospheric error has spatial correlation, applying a low-pass filter to the differential interferogram can isolate only the atmospheric error, allowing the extraction of the height difference component from the existing DEM. By adding this back to the original existing DEM, a high-precision refined DEM can be generated. The advantage is that the spatial resolution of the resulting height difference or refined DEM does not depend on the existing DEM's resolution but on the InSAR resolution, allowing for a high-resolution DEM even if the existing DEM's resolution is low.

This method was applied to two independent one-day interval pairs: December 23–24 and December 27–28, 2025. The observation mode was Staring Spotlight 4, with an incidence angle of 45°. Both pairs had a perpendicular baseline of about 500 m and a Height ambiguity of about 14 m. As a result, topographic changes exceeding 20 m around the crater due to the eruption were revealed with a spatial resolution of less than 1 m. The difference between the two independent pairs was generally within a few meters.

This case study demonstrates that high-frequency, high-resolution observation by small SAR satellite constellations is effective for assessing crater shape and topographic changes associated with eruptive activity, contributing to the advancement of future volcano monitoring methods.

Acknowledgements

The CR validation utilized deliverables from the FY2025 Demonstration Project Led by the Cabinet Office to Expand the Use of Small SAR Satellite Constellations. The installation and measurement of the CRs were carried out in cooperation with Pacific Consultants Co., Ltd.

Agriculture covers about 44% of the global habitable land and plays a particularly important role in global sustainability. Cropland makes up a third of agricultural areas and is crucial for food security (direct use and livestock feeding), raw materials, and biofuels. Livelihoods and economies around the world depend on agriculture and its close connection to the water and carbon cycle. About 100 million hectares of productive land per year are lost to land degradation, desertification, urbanization, or drought. Extreme weather such as droughts, floods, wind, and hail are magnified by climate change and endanger farmers around the world. Additionally, the uncertainties in agricultural carbon stocks and fluxes remain significant.

The airborne AGRIROSE-L campaign coordinated and performed by the German Aerospace Center (DLR) was conducted in cooperation with the LMU, GFZ and CzechGlobe over an agricultural area in southern Germany called Puch. The campaign is an ESA supported campaign. The campaign's primary goal is to provide calibration and validation data to support future Earth observation missions, specifically ROSE-L and CHIME, with a focus on improving the monitoring of soil moisture and health, crop growth, and other agricultural parameters from space. The data collected is crucial for developing and testing algorithms for sustainable agriculture.

ROSE-L mission will increase the understanding of soil moisture and agricultural systems with its high sensitivity to both vegetation and soil water. AgriROSE-L campaign aims to contribute to this understanding with the following goals:

• Imaging of a representative region of interest with varying crop types, crop stages, and soil moisture conditions every, at minimum, six days to ensure a high temporal resolution;
• Provision of an extensive time-series covering the full crop cycle;
• Simultaneous recording of C-Band and L-Band frequencies maximizing comparability;
• Gathering of tomographic, fully polarimetric, and interferometric information within a multi-baseline repeat-pass configuration. In addition, a number of headings has been flown;
• Acquisition in sub-meter azimuth and range resolution depending on the frequency;
• Acquisitions with three-day separation for higher temporal resolution and three flights per day for intra-daily information;
• Gathering of in-situ reference data on more than 10 fields in cooperation with the Ludwigs-Maximilians University, Munich;
• Acquiring data in the optical and hyperspectral domain for cross-comparison
• Providing the collected data as AI-ready datasets stemming from the multispectral as well as the SAR domain;
• Operation of the ground-based wideband scatterometer WBSCAT during the campaign
• Organization of acquisitions of relevant radar satellites to improve model transferability.

For this the DLR’s F-SAR system recorded a globally unique dataset across four different frequency ranges (the X, C, S and L bands). In total, the radar team carried out 23 measurement flights between April and July covering the whole agricultural vegetation season [1]. On selected days, the flights took place in the morning, at midday and in the evening to record any daily changes in the soil and vegetation.

The data was collected using innovative imaging techniques such as polarimetry, interferometry and tomography. Experienced DLR test pilots flew specified paths with metre-level precision, supported by the satellite-based navigation system integrated into the F-SAR. Parallel to each flight, a team from LMU collected ground measurements of soil and vegetation parameters, such as soil moisture, surface roughness, plant water content and plant biomass. In addition, four times also the DLR’s hyperspectral sensor HySpecs was flown over the same area and one-time CzechGlobes hyperspectral sensor acquired data.

In this research work the campaign, its collected data and the first performance analysis are presented.

[1] Campaign Implementation Plan, submitted to ESA 2025

SAR instruments have a long history in planetary exploration, particularly with regards to Venus. Venera 15&16 and Magellan have been exceptional past examples, making available the first Venusian Digital Elevation Model (DEM). Following their legacy, NASA’s VERITAS mission is being planned. In this context, the limited downlink capacity has long been a major constraint for Synthetic Aperture Radar (SAR) and Interferometric Synthetic Aperture Radar (InSAR) observations, making the development of efficient on-board compression strategies a critical aspect for such missions.

In this work, we propose a Deep Learning (DL)-based approach to jointly denoise and compress the interferometric phase (InSAR phase) on board. To this purpose, a Convolutional AutoEncoder (CAE) is trained end-to-end in a supervised manner and using a synthetic dataset, derived by exploiting experimental TanDEM-X data together with corresponding InSAR geometries and underlying topography. Such a network is naturally divided in two parts: an encoder and a decoder; the former, responsible for the compression of the acquired data, is supposed to be implemented on board the satellite, while the latter allows for decompressing and reconstructing the received data on ground. The encoder reduces the spatial dimensions of the input by applying a combination of convolutional and pooling operations, therefore achieving data compression as the dimensionality of the transmitted latent space is considerably lower than that of the original uncompressed data. On the other hand, the decoder reconstructs a denoised version of the original SAR interferogram from the received latent space on ground; however, as the original sampling is reconstructed through standard up-sampling techniques, the original high-resolution spatial details are partially lost. The proposed architecture is designed in a parametric way, thus allowing for hyper-parameter tuning on different settings, such as network depth and number of channels/features in the resulting latent space. A family of models is then trained targeting three different performance metrics: denoising capability, compression ratio, and preserved details.

The performance of the proposed methodology is assessed according to the three metrics mentioned above. The standard deviation of the phase error between the reference noiseless and estimated phase and across the test scenes is used as indicator for the denoising performance. The achieved compression ratio is evaluated using the Bits Per Original Pixel (BPOP) metric, representing the number of bits used by the compressed latent space with respect to the original number of pixels of the input uncompressed data. Last, the level of detail preserved in the autoencoder output is assessed by progressively removing high-frequency content from the reference noise-free phase and comparing the result with the decoder reconstruction; the best match determines the percentage of high-resolution details retained.

Results are then assessed with respect to a combination of boxcar filtering and JPEG 2000 compression, referred to as baseline method, which reflects one of the possible strategies reported in the literature and possibly to be exploited for currently-planned missions. As far as phase denoising is concerned, we employ a boxcar filter, consisting of an averaging operation over a sliding square window; more sophisticated filtering approaches, such as nonlocal filters, are not considered given their high computational complexity, which jeopardizes their use for extraplanetary missions. Note that denoising is achieved at the cost of spatial resolution, thus directly impacting the obtained compression rate as well. In particular, different windows are considered in order to investigate different trade-offs between denoising and compression performance.

As expected, boxcar filtering achieves a better denoising at the cost of spatial resolution. Indeed, when increasing the window size, the phase error decreases, while, on the other hand, less resolution is preserved. Also, bigger windows lead to higher compression ratios. As far as the proposed methodology is concerned, we assess different groups of models, each tuned on a specific performance metric. It is worth mentioning that the proposed autoencoder allows for a higher overall flexibility compared to the baseline method, since the whole performance space can be explored by tuning specific hyper-parameters during training. For example, a better denoising is expected at the cost of data volume reduction; also, in these cases, the corresponding models preserve more original high-resolution details compared to other autoencoder solutions, except for those optimized for this specific task. On the other hand, the best spatial details preservation is achieved at the cost of phase denoising capability, similarly to, e.g., a [3 x 3] boxcar filter. Note that the proposed autoencoder allows for an overall good trade-off between the three performance measures, even if tuning the network for the compression task only. Moreover, we evaluate some additional models representing good trade-offs between all performance measures. Overall, these solutions demonstrate competitive performance with respect to the considered baseline approaches.

As final remark, results look promising and have shown already an enhanced overall flexibility with respect to state-of-the-art baseline methods, in terms of performance scalability. Moreover, the proposed technique represents a valid alternative to most traditional approaches in terms of absolute performance as well, mainly targeting denoising capability and high-resolution details preservation.

During the past several years, we have designed, implemented, and operated compact, lightweight synthetic aperture radar (SAR) end-to-end systems at L- and S-band [1] for a broad range of applications. Our FMCW SAR systems are deployable on ground vehicles, UAVs, manned aircraft, and potentially future high-altitude pseudo-satellite (HAPS) platforms. They support SAR data acquisition and interferometric applications, including ground deformation monitoring, vegetation and forest parameter retrieval, and snow property estimation, etc. Our end-to-end system comprises: (1) a compact FMCW SAR sensor (L-band with up to 200 MHz bandwidth; S-band with up to 400 MHz bandwidth), (2) a compact GNSS-aided INS navigation unit (Honeywell HGuide n580/n500), (3) SAR focusing and processing software, and (4) interferometric processing and higher-level value-adding software tools. The FMCW-SAR architecture incorporates two alternating transmit channels and up to four simultaneous receive channels. Depending on mission requirements and platform constraints, the system supports single-polarization interferometry, polarimetric interferometry, and single-pass multi-baseline acquisitions. The system’s compact hardware form factor and a flexible SAR processing chain enable deployment on agile platforms (e.g., UAVs, vehicles) as well as platforms with stringent payload limitations (e.g., UAVs, HAPS). Consequently, a wide range of acquisition geometries and operational scenarios can be realized. Of particular relevance are use cases requiring short repeat-pass intervals and adaptable viewing geometries that complement spatiotemporal sampling schemes of current spaceborne SAR systems. Terrestrial, UAV-borne, airborne, or HAPS-based acquisitions enable quasi-geostationary repeat-pass observation schemes with short temporal baselines. Such configurations are well suited for monitoring rapidly evolving landslides, providing timely snow parameter updates, frequently mapping tropospheric water vapor distributions, and supporting emergency or disaster response scenarios. In our presentation, we include (1) results from repeat-pass interferometric campaigns conducted in 2024 and 2025 using L- and S-band Gamma SAR systems mounted on vehicles, including case studies of the Brinzauls (Switzerland) and Madesimo (Italy) landslides within the ESA MODULATE project, (2) results from subsequent UAV-based deployments using quadcopter and octocopter platforms, including interferometric analyses derived from repeated overflights.

In addition, UAV-based bistatic SAR imaging and interferometry provide a attractive and flexible framework for experimentally investigating bistatic SAR concepts.

At present, UAV-based bistatic SAR demonstrator platforms that are suitable for experimentally advancing bi-static and multi-static SAR mission concepts are only about to emerge. This applies to both key areas: (1) technology development—such as synchronization, precise positioning, and constellation design—and (2) UAV-based investigation of bistatic radar backscatter signatures across different bi-static geometries, including the development of new or improved bio- and geophysical parameter retrieval methods.

In 2025, we performed bistatic SAR experiments using two Gamma L-band SAR systems, each deployed on a Harris Aerial HX8 UAV. In our presentation, we include bistatic SAR imagery and interferometric results derived from these most recent experimental campaigns.

REFERENCES

[1] Frey, O., Werner, C., Leinss, S., Batt, T., Caduff, R., Dixon, T., Sadeghi Chorsi, T., Van Alphen, R., Schmitt, M., Eitel, M., Sica, F., Deeb, E., LeWinter, A., Filiano, D.L., Wagner, C.J., Hoppinen, Z., 2025: Multicopter-UAV- and car-borne repeat-pass SAR interferometry and SAR tomography with the compact Gamma SAR systems: first examples and use cases at S- and L-band, Proc. IEEE Int. Geosci. Remote Sens. Symp., Brisbane, Australia, 1374-1377.