KI4KI - Artificial Intelligence for Climate-Resilient Infrastructure Monitoring

KI4KI—Artificial Intelligence for Climate-Resilient Infrastructure Monitoring—was funded between 2022 and 2025 as a collaborative project of the BMWK (Federal Ministry for Economic Affairs and Climate Action) between Friedrich Schiller University Jena and the Ruhrverband (Grant ID: 50EE2202A). We developed AI-based approaches for multi-temporal interferometric synthetic aperture radar (MT-InSAR) time series to monitor deformations on dams in Germany.

Contents

Research Focus in Detail

  • 1. Suitability Analysis for a Satellite-based Monitoring Strategy

    In the first part of the project, the suitability of various types of dams (e.g., dam walls and embankment dams) for a satellite-based observation strategy was assessed. The methodology aimed to develop two indices to...:

    1. assess the general visibility of dams in radar imagery based on data availability, topography and land cover, and
    2. convert and interpret deformations in the sensor’s line of sight (LOS) into deformations in the radial direction relative to the crest of the structure.

    For this purpose, freely available MT-InSAR datasets from the German Ground Motion Service (BBD)External link were used.

    Ascending LOS sensitivity (%) to radial deformations in dependence on the radial orientation of the dam crest (°). The LOS sensitivity is expressed as a proportion of the true radial deformation (mm). Orthophoto: GDI-NRW (2024), EPSG: 25832.

    Image: Jonas Ziemer
  • 2. Combining Sensors of Different Wavelengths for Spatiotemporal Densification of MT-InSAR Time Series

    Segmentation of a dam into segments with different deformation characteristics. The PS points of the segments are shown in different colors and combined into a mean deformation profile over time.

    Image: Jonas Ziemer

    In the second step, data from two different radar sensors operating at different wavelengths – TerraSAR-X (X-band) and Sentinel-1 (C-band) – were combined to generate a spatially and temporally densified time series. To this end, a new tool called TSX2StaMPSExternal link was developed, which enables high-resolution TerraSAR-X data to be pre-processed using ESA’s open-source SNAP software. TSX2StaMPS is already integrated into snap2stampsExternal link as an updated version and was presented at ESA’s Fringe Conference in Leeds in 2023.

    The dams were divided into different segments and their deformation behavior was analysed using the combined TSX/S1 time series. By combining data from both sensors, it became possible to observe how the mean deformation profiles differ depending on the location of the individual segments. Furthermore, the deformation drivers were identified using in-situ data.

  • 3. Deformation Prediction using Data-driven Approaches

    As dam operators are concerned not only with analyzing past deformations but, in particular, with predicting future movements at their dams, the FSU-EO and FSU-CVG groups aimed to investigate the prediction of dam deformations using data-driven approaches. The aim was to improve the predictive accuracy of present models. A wide variety of algorithms were employed, including time-series forecasting models, ensemble methods and deep learning approaches. By implementing the algorithms into existing workflows, the prediction accuracy was considerably improved for both PS data and in-situ measurements.

    Left: Pendulum line (orange) and PS points (blue) at the Möhne Dam. The deformation prediction (in mm) for the year 2020 was determined through a comprehensive model search. Right: Turquoise dots represent the training data, while gray dots represent the test data recorded by the pendulum system (Ruhrverband, 2021). The colored lines show the predicted deformation from the best-performing model in each search space. Orthophoto: GDI-NRW (2024), EPSG: 25832.

    Image: Jonas Ziemer
  • 4. Monitoring Dam Deformations based on Electronic Corner Reflectors (ECR)

    Electronic corner reflector (ECR) with a solar panel for power. Installed here on the crest of the Bigge Dam.

    Image: Jonas Ziemer

    Unfortunately, not all dams meet the necessary criteria for a satellite-based monitoring strategy. For example, unfavourable topographical conditions or vegetation-covered surfaces can affect the detection of PS points on dam structures. To enable these dams to be monitored nonetheless, six electronic corner reflectors (ECRs) were installed at five dams managed by the Ruhrverband as part of the project. Over an observation period of more than 2.5 years, the ECRs provided stable radar signals and enabled deformation time series with millimeter accuracy. The results were compared with in-situ measurements carried out by the Ruhrverband and confirm the general suitability of ECRs for dam monitoring.

  • 5. Development of an Interface for Visualizing Results for Dam Operators

    In order to translate the scientific findings of the research into operational practice, a web-based platform has been developed that is specifically tailored to the needs of dam operators. The application brings together ECR-based satellite measurement data, the indices used in Section 1, and the operator’s in-situ measurement data within a single, interactive interface. Time-series visualisations, direct comparisons of multiple measurement points, and data exports in commonly used data formats facilitate the seamless integration of the results into existing monitoring programmes. Data ownership remains with the respective operators, while automated processing routines handle the preparation of new data sets and ensure reproducible processing. The system is designed for adaptation for other infrastructure operators and for different types.

    Web service for visualizing deformations at dams. Thanks to its modular design, the service can be customized to meet the specific needs of each dam operator.

    Image: Jannik Jänichen
Notice

Please find further information and details about the project results at GitHubExternal link and on the project page of the RuhrverbandExternal link. 

Contact

Annett Habenstein

Chair of Earth Observation
Picture of Annett Habenstein
Image: Annett Habenstein
Link to download vCard
vCard
JenTower, Room 26S04, 26. Etage
Leutragraben 1
07743 Jena Google Maps site planExternal link
Further link
To website
  • FSU
  • RV
  • DLR
  • BMWK