Conference Agenda

.

Systematic acquisitions of the Sentinel-1 constellation have opened a new era in change detection monitoring. From a basic image pair analysis, we can move to time series detection to gain more profound insight We are faced with a multifaceted, often ambiguous, yet well-defined change matrix. In the realm of crisis management, this information landscape can be quite cumbersome. Our presentation does not address the issue of change detection from a technological standpoint, rather, it focuses on the information needs of end users. In addition, we provide an available methodological overview to clarify specific technological challenges. Our objective is to examine the possible uses of different technologies by using crisis mapping as an example.

The “Romanian Ground Motion Pilot Service for Sustainable Infrastructure” (RO-GMS) addresses the Earth Observation (EO) market in Romania with the primary objective of facilitating the operational uptake of Persistent Scatterer Interferometry (PSInSAR) by public authorities. By leveraging the Copernicus European Ground Motion Service (EGMS), the project delivers tailored, actionable monitoring services for critical local infrastructure.

A central component of this initiative is a customized, user-friendly web platform (https://pstool.terrasigna.com/) engineered to lower the barrier to entry for users with limited remote sensing expertise. The environment hosts specialized analytical tools, including an interactive transect tool that enables PSInSAR results analysis and animations across target areas. Additionally, an on-the-fly analysis module provides immediate stability assessments at the scale of individual infrastructure elements uploaded by the user.

To enhance data accuracy, the project introduces advanced methodological improvements. Firstly, the standard PSInSAR processing algorithm was refined to successfully detect and map quick, highly non-linear ground dynamics. Secondly, we integrated a novel thermal layer, utilizing thermal sensitivity models calculated from EGMS data to separate temperature-correlated displacements from actual structural motion.

The efficacy of these advancements is demonstrated through two distinct case studies. The first case study highlights a bridge subjected to large-scale dynamic shifts related to thermal motion. The thermal models estimated from long term EGMS data and local weather information can be used to improve phase unwrapping in short time monitoring projects on areas with large thermal motion.

The second case study examines a water dam that underwent rapid subsidence triggered by a significant drop in the reservoir's water level. The Sentinel-1 derived PSInSAR results for the water dam were rigorously validated against in-situ terrestrial measurements spanning from 2017 to 2024. This ground-truth validation relied on a dedicated geodetic monitoring system, comprising a crest subnetwork and markers installed to measure quarterly the behavior of the surrounding rocks, using high-precision total stations and leveling instruments. The strong correlation between these geodetic field campaigns and the satellite observations underscores the reliability of the RO-GMS platform as a robust, operational tool for sustainable infrastructure management.

Part of the presented results were obtained within the ESA project “Verification of innovative applications integrating national InSAR capabilities and the European Ground Motion Service” (2025 – 2026).

Man-made targets at sea, which are critical for maritime domain awareness, environmental monitoring, and security surveillance, primarily include ships, oil and gas platforms, and wind turbines. These objects are monitored using remote sensing tools, notably Synthetic Aperture Radar (SAR), which can operate regardless of weather or lighting conditions.

Apparently, their monitoring by SAR measurements is an easy task but it not at all true Such a misunderstanding, although widely accepted, is supported by the idea that man-made targets are large, and metallic providing a distinct backscattering with respect to the sea background. Even when they are large and metallic there are some target shapes and SAR observation configurations that make them almost invisible. Of course, their detection is much more complex if even such two physical hypotheses are violated.

With the advent of artificial intelligence (AI), a large body of literature has focused on such applications. In this paper we like to present a critical analysis of such a problem describing the key physical issues that make it challenging. Further, a critical analysis of the reference SAR data set popularly used in these AI studies.

The study is completed by analyzing a physical-based SAR approach that makes benefit of intrinsic Maxwellian properties and polarimetric measurements.

Some first experiments regarding the Persian Gulf will be presented.

Persistent Scatterer Multi-Temporal InSAR (PS MT-InSAR) provides displacement time series after the initial interferometric processing stages, including coregistration, interferogram formation, phase processing, and mitigation of atmospheric and orbital effects. Temporal phase unwrapping remains one of the main sources of uncertainty in this framework, especially when deformation signals show nonlinear trends, rapid variations and strong noise. Under these conditions, conventional approaches based on simplified phase models may become unreliable, because they impose a temporal evolution that may not reflect the actual signal evolution. This work presents a new adaptive temporal phase-unwrapping method inspired by the tracking principle of a Phase-Locked Loop (PLL).
The PLL-inspired strategy follows the local phase evolution instead of forcing the data into a rigid global model. By combining adaptive tracking with a minimum-path criterion, the method selects physically plausible unwrapping solutions while preserving sensitivity to genuine temporal variations. It also remains computationally efficient, which allows it to scale to large MT-InSAR datasets without losing robustness on complex signals.
The proposed method can work as a standalone solution, but it can also check and refine existing results. In particular, it can be used in post-processing on displacement time series that have already been generated, as a quality check and as a means to improve the series themselves. This time-domain strategy complements existing model-based approaches. In other words, the two methods look at the same unwrapping problem from different and largely independent perspectives. This difference has practical value: when both methods agree, confidence in the solution increases; when they disagree, they highlight time series that likely need closer inspection.
The method was tested on real Sentinel-1 Persistent Scatterer time series covering both regional-scale and highly nonlinear deformation scenarios. The results show very high unwrapping accuracy where reference solutions are available, and they also show clear practical improvements on difficult nonlinear signals, where standard approaches more easily fail or become unstable. Overall, the method combines accuracy, flexibility, and complementarity with existing approaches, which makes it a promising tool for advanced PS MT-InSAR time-series analysis and deformation monitoring in complex real-world conditions.

Monitoring refugee camps is key for UNHCR, the United Nations Refugee Agency, and other humanitarian organisations. Whether formal or informal, they can greatly vary over time in size and population. Ensuring proper food and water supplies, as well as meeting sanitary needs, requires adequate knowledge of the population and its distribution.

Registration systems provide the most reliable source of population data, but they do not always capture subtle or sudden changes in population dynamics. Furthermore, such systems are more commonly implemented in formal camps and are less prevalent in informal camps.

We present a complementary method using satellite time-series imagery from ESA Sentinel-1. The latter, being free, with worldwide coverage and a regular revisit, is particularly adapted to this monitoring. Refugee camps can be made of many different materials, from standardized tents to shelters made of local materials. These camps are being set up on bare soil; whatever the shelter type, the resulting radar backscattered signal is increased, especially at C-band, with the medium-resolution of Sentinel-1. The time series are first cleaned of speckle and ephemeral targets (e.g., vehicles, which can produce strong signals and falsify the analysis) using the Frozen Background Reference technique [1]. They are then filtered out for moisture anomalies, which would alter the analysis. We have then defined some metrics. They allow us to map the camp's density over time and track population variation.

Foliage Penetration Synthetic Aperture Radar (FOPEN SAR) operating at low frequencies offers unique capabilities to penetrate dense vegetation canopies. This allows for the detection of concealed man-made objects and underlying structural changes that are critical for applications like illegal logging detection and border surveillance in heavy forested terrains. However, target and change detection in FOPEN SAR faces significant challenges due to complex volume scattering from vegetation layers, non-homogeneous background clutter, and the inherent poorer spatial image resolution associated with longer wavelengths, all of which contribute to high false alarm rates.
Classical target and change detection algorithms in this domain have been dominated by Constant False Alarm Rate (CFAR) and its different variants, primarily due to computational efficiency and suitability for intensity-only datasets (e.g., CARABAS II VHF). While deep learning models such as Convolutional Neural Networks (CNNs), diffusion models, and transformers offer improved detection accuracy, most existing research is concentrated on C-band and X-band data. Translating these advanced models to longer-wavelength P-band/L-band data remains an open challenge. Such difficulty is driven not only by fundamentally different scattering mechanisms, such as complex sub-canopy volume scattering and severe signal attenuation, but also by the lack of large-scale, reliably annotated public datasets for training robust neural networks.
To address these limitations, we introduce a novel deep learning framework explicitly tailored for Long-wavelength SAR target and change detections. We pivot from classical amplitude-only techniques to methods that leverage both the amplitude and the phase information derived from (Geo-corrected) Single-Look Complex (SLC) image pairs. Specifically, we integrate the metric learning within a Siamese CNN architecture. Instead of relying on simple image differencing, the network is trained to project co-registered multi-temporal scenes into a highly discriminative embedding space. By explicitly optimizing a distance metric, the model learns to maximize the feature-space distance between potential target/change signatures and stable background clutter.
Experiments on the SAOCOM L-band dataset show the proposed algorithm effectively detects prominent changes, such as new infrastructure, and semi-occluded objects near forest edges. Additionally, it demonstrates strong potential for predicting completely obscured, sub-canopy changes within dense forests. Verifying these specific, deeply concealed detections through targeted field campaigns will be the primary focus of our future work once comprehensive ground-truth data becomes available.