Unsupervised Analysis of Large-Scale EGMS PS-InSAR Time Series for Ground Deformation Processes: A Case Study in Iceland

Earth surface deformation due to volcanic and tectonic activities, and landslides have significant impacts on both human society and the natural environment. Satellite remote sensing, particularly Interferometric Synthetic Aperture Radar (InSAR), is a powerful tool to obtain extensive ground displacement spatio-. However, at large spatial scales, the monitoring data often contain mixtures of multiple deformation processes, making direct interpretation highly challenging. This complexity calls for data-mining approaches to make large-scale ground motion monitoring data readily interpretable for end users.

To address this issue, we investigate an unsupervised and explainable framework for large-scale analysis of InSAR displacement time series. We use the European Ground Motion Service (EGMS) Level 2b ascending and descending dataset over Reykjanes Peninsula, Iceland, as a case study. The proposed workflow integrates statistical source separation and deep learning-based clustering to extract, group, and interpret dominant deformation patterns.

First, a statistical analysis of large scale InSAR time series is conducted using independent component analysis to extract the dominant deformation sources. Second, these components are interpreted in terms of physical processes by integrating external geophysical and environmental datasets, such as geological maps, tectonic structures, and topographic features. Third, a deep clustering network is applied to the time series data to group deformation patterns into interpretable categories that reflect distinct ground motion behaviours.

This work contributes towards the development of scalable and explainable ground motion classification analysis tools from massive InSAR time series data, offering valuable support for decision-making and early warning systems in relevant management and disaster response agencies.