Detection and classification of large-scale ground motion from remote sensing data: A case study in Hesse, Germany

Abstract

Large-scale subsidence and uplift pose a significant risk to buildings and infrastructure. While subsidence due to groundwater removal or construction activities can easily be constrained on a local scale, regional changes caused by climate change are more difficult to detect. These phenomena are investigated within the „Umwelt 4.0, Cluster I - Use of digital terrain models and Copernicus data" project, which is carried out by the Hessian Agency for Nature Conservation, Environment and Geology in cooperation with the TU Darmstadt and funded by the Hessian Minister for Digital Strategy and Development. Within the framework of this project, we are creating a systematic workflow to detect ground motion over a period of several years. We focus on the state of Hessen, Germany, where several regions are known for landslide activity, e.g., Hoher Meissner, or for widespread subsidence, e.g., in the industrial areas surrounding Frankfurt a.M.. In this way, occurring ground movements and even mass movements could be detected at an early stage and, if necessary, measures can be initiated. Based on these results, future decisions on regulations or even information for the general public on risk areas can be created.We utilize two major datasets based on remote sensing. High-resolution digital elevation and surface models (DGM 1 and DSM 1) from airborne LiDAR surveys by the Hessian Administration for Land Management and Geoinformation. For the most parts of Hessen, it was possible to calculate differences in elevation between the years 2014, 2019 and 2021. The second dataset are persistent scatterer interferometry points (PSIs) from the BodenBewegungsdienst Deutschland with a temporal resolution of 6 days since 2015. Both datasets are integrated and linked with other data sources, such as geological maps, known subsidence-sensitive layers, hydrogeological and climatic data. For the InSAR data a toolbox has been developed that automatically detects regions with strong movement (Ground Motion Analyzer). A major challenge for integrating both datasets is the large difference in spatial coverage and temporal resolution. Advantages of LiDAR data are the high spatial resolution and the possibility to detect even small-scale movements (<5 x 5 m) below vegetation cover, for example the re-tracing of forest roads or the creation of logging trails. A disadvantage is the low temporal resolution of several years between flights in comparison with the 6 days of the PSI data. From the latter, even seasonal variations can be detected and measured. However, the spatial distribution of the points is highly heterogeneous, so in cities the point density is very high, whereas in rural areas hardly any measurements exist. Other problems are the strong fluctuations both within a time series of a single PSI point and between neighbouring points.With our contribution we want to highlight a typical use case of both data sets and their implementation into regulatory decision-making processes. Furthermore, we want to show a possible integrative method combining remote sensing data with ground based geoinformation and future use of advanced classification schemes to automatically detect affected regions in big datasets.