Conference Agenda

In this contribution, we will focus on two main developments supporting the case for structural health monitoring of infrastructure, in particular railway structures.

First, we developped a postprocessing framework designed to improve the fusion of multitrack InSAR (Interferometric Synthetic Aperture Radar) point clouds for urban deformation monitoring. This addresses two major challenges in integrating SAR datasets from different satellites and viewing geometries: inconsistencies in geometric height references and mismatches in displacement datums.

The proposed framework introduces a dual-correction strategy. First, it performs geometric datum correction using iterative tie-point identification and least-squares adjustment. Corresponding persistent scatterer (PS) points from different tracks are identified after coordinate standardization and facade filtering. Elevation offsets between tracks are then estimated and iteratively refined to align all point clouds into a unified three-dimensional reference frame. Second, GeoDisFusion corrects temporal displacement inconsistencies by aligning deformation velocities and time-series references between tracks, ensuring consistent displacement measurements across datasets.

The framework was validated over Kunming, China, using four SAR datasets with different imaging geometries: Sentinel-1 ascending and descending tracks, TerraSAR-X descending data, and ALOS-2 PALSAR-2 ascending data. Results showed substantial improvements in both spatial and temporal consistency. Geometric correction reduced median elevation discrepancies from approximately 15 m to 5 m when compared with airborne LiDAR reference data. In urban test regions such as Yunnan Artist Park and Changshui Airport, corrected PS points aligned much more accurately with building outlines and infrastructure features.

For deformation analysis, the method reduced inter-track deformation velocity differences by up to 1.6 mm/year, significantly improving agreement between datasets. The fused point cloud also achieved a higher effective PS density than any individual dataset, enhancing spatial coverage and deformation detection capability. Overall, the method (GeoDisFusion) provides a robust and geodetically consistent framework for integrating heterogeneous multitrack InSAR observations, enabling more reliable urban deformation monitoring without requiring external geodetic reference data.

Second, we developed Instantaneous State InSAR (ISI), a new framework for near real-time displacement monitoring using sequential InSAR observations. Unlike traditional InSAR time-series approaches, which rely on batch processing and estimate static parameters over an entire observation period, ISI models the instantaneous kinematic state of a scatterer, enabling continuous updates as new SAR acquisitions become available.

The proposed method combines recursive least-squares estimation with a Kalman-filter framework. Instead of assuming constant displacement behavior, the method estimates instantaneous position and velocity states and updates them sequentially with each new observation. A key innovation is the modeling of velocity as a temporally correlated Ornstein–Uhlenbeck process, which introduces physically meaningful smoothness constraints into the displacement signal. These constraints are controlled by two parameters: the velocity standard deviation and the decorrelation time, which together regulate the magnitude and smoothness of the estimated motion.

The methodology begins with a static initialization stage using integer least-squares ambiguity resolution and conventional batch estimation. After initialization, recursive state prediction and measurement updates are performed whenever a new SAR acquisition becomes available. The framework also incorporates the normalized median absolute deviation (NMAD) of the amplitude as a proxy for phase quality, allowing adaptive stochastic modeling of observations. Importantly, the smoothness constraints enable implicit phase unwrapping by ensuring that predicted phase residuals remain within half a wavelength cycle.

The method was evaluated using nine years of Sentinel-1 data. Results show that ISI achieves estimation quality comparable to full batch methods in stable conditions while significantly outperforming them in dynamic situations. The method better captures abrupt changes, anomalies, and evolving displacement behavior, whereas conventional batch approaches often smooth out or miss these events entirely. Examples demonstrated improved adaptation to displacement changes, reduced cycle-slip effects, and more realistic tracking of temporal dynamics.

Overall, ISI provides a computationally efficient and flexible framework for near real-time infrastructure and urban stability monitoring. By combining recursive updates, dynamic state estimation, and physically motivated smoothness constraints, the approach improves responsiveness to changing ground motion while maintaining robust estimation accuracy.