Analysis of ground deformations based on parallel geostatistical computations of PSInSAR data

The ground deformations can be caused by many different factors. One of them is coal exploitation. It can cause not only strong and abrupt ground movements but also small, long period deformations which can occur even several years after exploitation is finished. In both cases these displacements are serious menace for surface and subsurface infrastructure. In this work the PSInSAR data were used to study small, long lasting ground deformations which occurred in the Upper Silesian Coal Basin (south Poland) in the years 1992–2003. In this region the intensive coal exploitation has been carried on. Additionally this area has complicated geological structure. These factors make studied region particularly threatened with terrain deformations. The subsidence phenomenon was detected in this region using PSInSAR method. PSInSAR technique is a dynamically developed branch of satellite radar interferometry. It exploits a set of dozens satellite SAR images in order to detect small ground deformations for large areas. PSInSAR technique derives information only about ground movements for stable radar targets, so called PS points. They correspond with man-made features on the ground like buildings, bridges etc. In the Upper Silesian Coal Basin the values of ground deformations were measured for more then 120000 PS points. Their location is very irregular. In order to study the origin of this ground displacements the values of deformations have to be estimated in no-observed locations. In this work this estimation was performed using kriging which is a geostatistical method of interpolation. It is very computationally expensive method because it requires the solution of a large linear system for each grid node in which the interpolation is done. Because the PSInSAR dataset is very large the kriging computations were done in parallel environment. In the designing of the parallel kriging algorithm also strong irregularity of PS points location was taken into consideration. At the beginning of the formulated parallel kriging algorithm, the number of PS points used in grid nodes interpolation was determined. Then, based on this information, data decomposition was conducted. In this process, grid points were divided between CPUs, so that each processor had the same number of calculations to perform. This solution is optimized for PSInSAR data points, which are deployed irregularly, often in local centers on the research area. It was shown in this work that geostatistical methods can be successfully used in PSInSAR data analysis, but are computationally expensive. By designing parallel kriging algorithm for PSInSAR data and using Blade infrastructure for computations, it was possible to conduct the interpolation of ground deformations in viable time. It is worth to underline the universality of this algorithm, which can be used not only for PSInSAR data, but also for other types of data with similar characteristic.