Python Tool Analyzes Continental Radar Ground Motion Time Series Across Europe

Researchers have developed an open-source software tool called EGMS-PCA that automates the temporal statistical analysis of continental ground deformation time series collected across Europe by the European Ground Motion Service. Operating as a component of the European Union Copernicus Land Monitoring Service under the direction of the European Commission in cooperation with partners such as the European Space Agency, the ground monitoring system produces millimeter-precision measurements of surface displacements derived from multi-temporal Interferometric Synthetic Aperture Radar processing of Sentinel-1 satellite radar imagery. The newly introduced Python application couples an interactive graphical interface with Principal Component Analysis and K-means clustering algorithms to automatically ingest, align, filter, and partition massive volumes of satellite observations into distinct, interpretable displacement patterns without requiring programming expertise from users.

The underlying satellite data originate from the European Union Copernicus Earth observation program, which maintains spaceborne Synthetic Aperture Radar constellations transmitting electromagnetic waves in a right-looking geometry inclined at look angles between 29 degrees and 46 degrees from nadir. Earth surface observations are captured in the satellite line-of-sight across polar orbits traveling either northward in ascending tracks or southward in descending tracks. The Sentinel-1 mission operates in an Interferometric Wide Swath mode covering 250-kilometer swaths across three sub-swaths at a full spatial resolution of 20 by 5 meters. Sentinel-1A began operating in April 2014, while Sentinel-1B provided complementary coverage from April 2016 until an electronics power supply anomaly halted data delivery in December 2021. Continuity of the 6-day orbital revisit cycle was subsequently restored with the launch of Sentinel-1C in December 2024 and Sentinel-1D in November 2025, maintaining small spatial and temporal baselines ideal for differential interferometric processing.

Ground surface motions within the European Ground Motion Service are computed through multi-temporal interferometric radar processing implemented via an Integrated Wide Area Processor that utilizes algorithms such as Permanent Scatterer Interferometry, Small Baseline Subsets, SqueeSAR, PSP-IFSAR, and GSAR-GTSI. The service distributes measurements through an online explorer across three processing levels: basic Level 2a products providing uncalibrated line-of-sight displacement velocities relative to local reference points, calibrated Level 2b products corrected via regularized Global Navigation Satellite System velocity models to establish absolute displacements, and ortho Level 3 products. Level 3 data combine ascending and descending calibrated observations onto a standard 100-meter grid in the EPSG:3035 coordinate reference system, resolving absolute horizontal east-west and vertical up-down motion time series as virtual grid cells representing local average terrain motion.

Handling these tabular data files presents significant computational bottlenecks for non-specialist researchers, as continental updates following five-year moving windows comprise approximately 15 billion measurement points spanning roughly 10 terabytes of data. While standalone community tools such as EGMStream and EGMS-toolkit were previously introduced to assist with spatial searching, downloading, and merging operations, interpreting the resulting time-series arrays traditionally required specialized statistical scripting. Previous scientific approaches for interpreting interferometric time series ranged from predefined linear, polynomial, or wavelet models to blind source separation through Independent Component Analysis and supervised deep learning networks. The EGMS-PCA framework directly operationalizes an unsupervised framework originally described by Festa and colleagues, transforming an abstract analytical workflow into an automated, distributable software package that natively processes multi-tile tabular ground motion files.

The software architecture automates data ingestion, checks user declarations against internal file naming conventions and track angles, and harmonizes heterogeneous temporal acquisitions across adjacent satellite footprints. When neighboring data tiles possess disparate start dates or missing acquisition epochs, an adaptive timeline alignment algorithm detects the maximum overlapping temporal window and applies Piecewise Cubic Hermite Interpolating Polynomials. This mathematical formulation reconstructs missing values while preserving original signal dynamics. The software generates diagnostic cross-validation plots displaying root mean squared error and Pearson correlation coefficients against raw observations. Users can interactively specify minimum and maximum velocity filtering thresholds or supply Keyhole Markup Language boundary files to spatially clip the ingested points to specific regions of interest, recording all configuration parameters in a standardized reproducible text log before initializing the multivariate pipeline.

To perform temporal pattern discovery, the software standardizes the aligned time series across measurement points for each acquisition date, eliminating mean shifts and scaling to unit variance to ensure equal statistical weighting across all temporal steps. Principal Component Analysis is then executed to capture the dominant temporal deformation modes, with the optimal number of principal components automatically identified through an automated knee-point detection algorithm evaluating the inflection point of the explained variance curve. Rather than clustering transformed component scores, the tool applies a Euclidean Mini-Batch K-means clustering algorithm directly to the aligned time series in millimeters, maintaining physical displacement magnitudes so that resulting clusters represent real-world annual deformation rates. To prevent workstation memory exhaustion when clustering datasets comprising hundreds of thousands to millions of points, execution is partitioned into multithreaded processing blocks of 8,192 rows using vectorized matrix calculations.

Researchers validated the automated workflow through a regional benchmark covering Northern Ireland using six Level 3 ortho vertical displacement tiles spanning January 13, 2019, to December 18, 2023. The ingested dataset initially comprised 981,283 measurement points extending across portions of Ireland and Wales. Applying a regional boundary polygon reduced the sample to 382,440 measurement points across a standardized timeline of 301 temporal acquisitions. Principal component decomposition revealed that the primary component accounted for 70.48 percent of total dataset variance, while the second component captured 6.72 percent, followed by 2.71 percent for the third and 1.45 percent for the fourth. The automated inflection detector selected five principal components, thereby establishing five discrete clusters that converged mathematically in 83 iterations of the mini-batch routine.

The five identified clusters exhibited clear separation by mean vertical displacement rates across Northern Ireland: cluster UP-1 exhibited pronounced downward displacement averaging minus 8.5 millimeters per year, cluster UP-2 averaged minus 5.1 millimeters per year, cluster UP-3 reached minus 2.1 millimeters per year, cluster UP-4 remained substantially stable at minus 0.3 millimeters per year, and cluster UP-5 showed slight upward movement at positive 0.7 millimeters per year. Benchmarking on an Intel Core i9-14900K workstation running Windows 11 with 128 gigabytes of RAM completed the entire pipeline—from multi-tile ingestion and polynomial interpolation to principal component decomposition, clustering, diagnostic visualization, and export of EPSG:3035 shapefiles—in 1,445.59 seconds, or approximately 24 minutes, with a peak memory footprint of 12.0 gigabytes.

To evaluate physical correspondence, the authors compared the spatial distribution of the resulting clusters against independent Corine Land Cover peatland boundaries using point-in-polygon spatial queries across all 382,440 points. The proportion of points situated within mapped peatland environments increased monotonically with downward displacement velocity, rising from 8.1 percent in cluster UP-5 and 5.4 percent in stable cluster UP-4, to 37.5 percent in cluster UP-3, 63.9 percent in cluster UP-2, and 67.6 percent in cluster UP-1. Jointly evaluating the two subsiding clusters as an indicator of mapped peatland yielded a precision of 0.65, recall of 0.40, an F1-score of 0.49, and a Cohen’s kappa coefficient of 0.43 against the 1:100,000 scale land cover reference.

The study authors emphasize that these statistical associations reflect physical subsidence driven by the oxidation and compaction of drained organic soils, alongside seasonal soil moisture swelling and shrinkage verified through Sentinel-2 Normalized Difference Moisture Index imagery. However, the investigators caution that the automated clusters represent mathematical groupings of similar temporal trajectories rather than direct diagnoses of specific geological mechanisms. Furthermore, nationwide analyses can suppress subtle localized kinematic anomalies due to macro-regional variance dominance, although targeted spatial clipping allows local high-amplitude seasonal cycles to re-emerge. Available under open-source distribution on Zenodo with complete dependency specifications, the Python toolkit offers territorial planners, geoscientists, and public monitoring authorities an accessible mechanism for rapid, reproducible preliminary screening of large-scale ground motion data.

Subject of Research: Automated temporal analysis of European Ground Motion Service satellite time series

Article Title: EGMS-PCA: a Python GUI tool for temporal analysis of EGMS time series

Article References: Poggi, F., Fibbi, G., & Raspini, F. (2026). EGMS-PCA: a Python GUI tool for temporal analysis of EGMS time series. Earth Science Informatics, 19(12), Article 245. https://doi.org/10.1007/s12145-026-02288-3

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02288-3

Keywords: EGMS-PCA, Copernicus Program, European Ground Motion Service, InSAR, Synthetic Aperture Radar, Sentinel-1, Principal Component Analysis, K-means Clustering, Ground Deformation, Time Series Analysis, Geographic Information Systems, Subsidence Monitoring, Peatland Deformation, Remote Sensing, Earth Observation, Python Scientific Software