Ground deformation associated with post-mining activity at the French–German border revealed by novel InSAR time series method

Abstract

We present a novel methodology for integration of multiple InSAR data sets for computation of two dimensional time series of ground deformation. The proposed approach allows combination of SAR data acquired with different acquisition parameters, temporal and spatial sampling and resolution, wavelength and polarization. Produced time series have combined coverage, improved temporal resolution and lower noise level. We apply this methodology for mapping coal mining related ground subsidence and uplift in the Greater Region of Luxembourg along the French–German border. For this we processed 167 Synthetic Aperture Radar ERS-1/2 and ENVISAT images acquired between 1995 and 2009 from one ascending (track 29) and one descending (track 337) tracks and created over five hundred interferograms that were used for time series analysis. Derived vertical and east–west linear deformation rates show with remarkable precision a region of localized ground deformation located above and caused by mining and post-mining activities. Time series of ground deformation display temporal variability: reversal from subsidence to uplift and acceleration of subsidence in the vertical component, and horizontal motion toward the center of the subsidence on the east–west component. InSAR results are validated by leveling measurements collected by the French Geological Survey (BRGM) during 2006–2008. We determined that deformation rate changes are mainly caused by water level variations in the mines. Due to higher temporal and spatial resolution the proposed space-borne method detected a larger number of subsidence and uplift areas in comparison to leveling measurements restricted to annual monitoring of benchmark points along roads. We also identified one deformation region that is not precisely located above the mining sites. Comparison of InSAR measurements with the water levels measured in the mining pits suggest that part of the water that filled the galleries after termination of the dewatering systems may come from this region. Providing that enough SAR data is available, this method opens new opportunities for detecting and locating man-made and natural ground deformation signals with high temporal resolution and precision.

Introduction

Synthetic Aperture Radar Interferometry (InSAR) is an established methodology for mapping ground deformation of natural (e.g. Schmidt et al., 2005, Beavan et al., 2010, Gonzalez et al., 2010, Chen et al., 2012, Zhao et al., 2012) and anthropogenic (e.g. Raucoules et al., 2003, Strozzi et al., 2011, Jiang et al., 2011, Zhang et al., 2012) causes. An interferogram is a conjugate product of two co-registered complex SAR images acquired by the same or similar sensor at two different times (Massonnet and Feigl, 1998, Rosen et al., 2000). After removing Earth's curvature and topographic components, the InSAR image (also called differential interferogram) measures the ground deformation that occurred between image acquisitions but also signals related to travel time delays caused by water vapor in the troposphere and fluctuations of the electron content in the ionosphere (Hanssen and Feijt, 1996, Li et al., 2005). In a single differential interferogram deformation and atmospheric signals are indistinguishable, which limits accuracy and applicability of InSAR in case of small to moderate ground deformation (e.g. Gonzalez et al., 2010, Samsonov et al., 2010, Samsonov et al., 2011a).

The ability of InSAR to map ground deformation with sub-centimeter precision over a large area with only a fraction of the cost of ground based measurements has stimulated rapid progress in methodology development. One of the first studies that introduced InSAR applied to Earth Observation was of Massonnet et al. (1993), in which a single differential interferogram over southern California captured co-seismic displacements of the 1992 M7.3 Landers earthquake. A few years later Sandwell and Price (1998) introduced a stacking technique in which interferograms were combined into a single product with improved signal-to-noise ratio. A few years later Ferretti et al. (2001) introduced persistent scatterers (PS) methodology that extended applicability of InSAR to densely vegetated regions that were otherwise interferometrically incoherent. Finally Berardino et al. (2002) and Usai (2003) introduced Small Baseline Subset (SBAS) methodology that produces time series of ground deformation by computing a least-square solution from a large subset of interferograms and Hooper (2008) combined PS and SBAS techniques into a single methodology. Various modifications were also presented (e.g. Samsonov et al., 2011b, Hu et al., 2012), including those developed for studying 3D deformation (Rocca, 2003, Wright et al., 2004).

Fast progress in development of advanced InSAR methodologies was also caused by a steady increase in data quantities available for InSAR analysis. In the early 1990s the only operational SAR satellite was ERS-1 launched by the European Space Agency. By the early 2000s there were a few operational satellites (ERS-2, ENVISAT from ESA, RADARSAT-1 from CSA and ALOS from JAXA). Thanks to the generous data policy of ESA and other space agencies large sets of SAR data became widely accessible, promoting further development. A decade later, in the 2010s, multiple SAR satellites and satellite constellations (RADARSAT-2, Cosmo-SkyMed, TerraSAR-X) produce ever-growing quantities of data that demand further progress in methodology development for maximizing the effectiveness of interpretation.

At present the widely applied processing methodology is based on the SBAS technique. However, the standard SBAS when applied in a current environment has a few limitations: (i) it can only handle one InSAR data set at the time; (ii) it produces only the line-of-sight deformation time series, which are hard to interpret in case of complex signal; (iii) the produced time series have limited temporal coverage and coarse temporal resolution; (iv) due to poor temporal resolution the advanced filtering or/and regularization techniques have very limited applicability; consequently, (v) due to the poor temporal resolution and low signal-to-noise ration transient signals cannot be properly detected.

All these issues are automatically resolved in the proposed Multidimensional SBAS (MSBAS) method. If InSAR data from more than one orbital geometry is available MSBAS produces: (i) the multidimensional time series of ground deformation, in case of space-borne InSAR, vertical and horizontal east–west components; (ii) all InSAR data sets are utilized simultaneously, which allows to achieve uninterrupted temporal coverage and dense temporal resolution; (iii) dense temporal resolution allows the advanced processing, such as filtering and regularization, which improves signal-to-noise ratio and allows detection of the low-amplitude short-duration transient deformation.

In Samsonov and d’Oreye (2012) we demonstrated that over the past decade, hundreds of SAR images were collected over Virunga Volcanic Province (VVP, the Democratic Republic of Congo) from various satellites with different wave-length, resolution, acquisition geometry, and temporal sampling. These SAR images combined together produce over a thousand differential interferograms subdivided into eight SBAS time series (for each individual subset). VVP is not an exceptional region and even larger sets of data have been collected for many other areas including those selected as the research community supersites (http://supersites.earthobservations.org/). Manual analysis and interpretation of large sets of data consisting of hundreds to thousands of interferogram is clearly beyond human abilities and, therefore, new approaches are warranted.

In Samsonov and d’Oreye (2012) we provided in depth theoretical derivation of the MSBAS technique and performed error and sensitivity analysis in case of the non-negligible north–south component. We produced two dimensional time series of volcanic ground deformation derived from eight InSAR data sets over the highly coherent lava flow areas in the high altitude regions subjected to the significant troposheric noise.

In this paper we apply a newly developed MSBAS methodology for mapping coal mining related ground subsidence and uplift in the low-coherent densely vegetated the Greater Region of Luxembourg along the French–German border (Fig. 1) and produce two dimensional time series of ground deformation derived from only two InSAR data sets. Historical leveling results reported in GIATM (2007) suggest that broader scale subsidence in this region started prior to 1961 (starting date of leveling campaigns). Rates of subsidence varied from a few mm/year to about 1 m/year but the area that experienced fastest subsidence according to leveling was only a few hundred m². The first InSAR results for this area were reported in Raucoules et al. (2007). The 1993–1993 ERS interferograms presented in Raucoules et al. (2007) had very limited coverage due to decorrelation, sufficient to identify presence of ground deformation but insufficient to measure the deformation rate precisely. The 1993–1994 JERS-1 interferograms showed localized areas with more than two fringes of LOS displacements (about 24 cm in total), however, such fast motion has not been confirmed by the ERS data nor the topographic error was evaluated, that could have been responsible for the observed fringes.

In this paper we combined over five hundred ERS-1/2 and ENVISAT interferograms from ascending and descending tracks to produce two components of the ground deformation with a remarkable precision. Ground deformation measured by our space-borne method is validated by comparison with the leveling measurements performed by the French Geological Survey (BRGM) during 2006–2008. Detected deformation rate changes are attributed to identified causes: termination of exploitation, termination of the dewatering operations and variation in the water levels in the abandoned galleries. At our resolution level we did not observe in our InSAR results any evidences of fast deformation but it cannot be excluded at the sub-resolution level. Using this technique it is possible to reconstruct the complete 3D motion when data from at least three, including one non-near polar orbiting (e.g. air-borne), sensors becomes available.