Drivers and spatio-temporal characteristics of land movements associated with sinkhole formation in flooded, abandoned mines

Flooding of underground mines after closure often leads to groundwater rebound, land uplift, and sinkhole formation. Such land surface changes create direct risks for infrastructure and for people living in affected areas. In recent years, the number of underground mines being closed has increased, particularly in Europe, which raises the need for efficient techniques that can monitor land surface deformation, improve process understanding in post-mining areas, and support the planning of mitigation and land-management measures.
This study focuses on the Olkusz-Pomorzany zinc-lead underground mining district (southern Poland), closed in 2020, where groundwater pumping stopped in 2021, and since then numerous sinkholes have been reported, together with uplift and local subsidence zones affecting forests, fields, and infrastructure. The area is a challenging case because mining has been conducted since the 13th century using different methods, including shallow and partly undocumented workings, and because the geological setting includes fractured and locally karstified carbonate rocks covered by Quaternary unconsolidated deposits of highly variable thickness.
We present land surface deformation results based on satellite radar interferometry and time-series analysis. We use Sentinel-1 Persistent Scatterer InSAR (PS InSAR) for 2021-2024 to map long-term motion patterns, and short-period SBAS results from Sentinel-1 and SAOCOM (2023) to better capture faster, non-linear changes associated with sinkhole formation in areas where C-band coherence is limited. The displacement time series are analysed in two ways. First, Independent Component Analysis (ICA) is applied to separate the main signals present in the whole dataset, including long- and short-term deformation patterns and acquisition-geometry effects; ICA is then used to map spatial differences in component amplitudes and to highlight zones with changing deformation dynamics. Second, each time series is classified into stable, linear, non-linear (quadratic), bilinear, and discontinuous behaviours, including sensitivity tests of the classification thresholds to provide reliable separation between stable and moving points.
Results are compared with a sinkhole inventory (2021-2024) and with mining layers, including the extent of historical workings, areas of shallow mining, and local geological conditions such as overburden type and thickness. Sinkholes are most often located within zones of strong uplift, associated with groundwater rebound, but they do not occur across all uplifting land surface. Instead, they cluster where the time series show the largest change in deformation rate (acceleration/deceleration or breaks) and where shallow, old mining is reported. These patterns suggest that rebound-driven uplift sets the regional background signal, while local weakening and structural complexity related to legacy workings control where sinkholes occur. The combined ICA and trend-classification approach helps to separate regional-scale rebound from local instabilities and supports targeted hazard management in post-mining landscapes.