Quantifying Earth Surface Change with Multi-sensor SAR Imagery
Bin Zhang is a PhD student in the department of Earth Observation Science. (Co-)Supervisors are prof.dr.ir. A. Stein and dr. L. Chang from the faculty of Geo-Information Science and Earth Observation.
Earth surface changes are closely related to the development of human society and the environment. To estimate changes in their shape and site, satellite Synthetic Aperture Radar (SAR) and Interferometric SAR (InSAR) have been developed. Thanks to the significant advantages of SAR imagery for the purpose of surface change monitoring, e.g., high spatial and temporal resolution, all-weather, all-time and high precision, this study concentrates on InSAR for natural and human-induced subsidence and SAR for tidal flats. Several issues restrict the applicability and feasibility of InSAR in surface change estimation. Firstly, dealing with big data poses new challenges for data fusion. Secondly, improper functional and stochastic model selection for phase unwrapping leads to biased parameter estimations. Thirdly, traditional InSAR deformation time series analysis methods are not always useful to high decorrelation and limited detection accuracy.
In this thesis, three major challenges were addressed. First, a spatio-temporal SAR data integration method was developed to address two issues: 1) identification of common ground targets from different SAR datasets in space, and 2) concatenation of time series when dealing with temporal dynamics. Geolocation uncertainty of InSAR Measurement Points (IMP) together with Monte Carlo methods have been used for SAR data spatial integration. Multiple Hypothesis Testing (MHT) was employed to distinguish the best deformation time series model. This model was used for SAR data temporal concatenation. The integration method was applied for monitoring surface deformation in the city of Rotterdam with medium and high resolution Radarsat-2 SAR data. The result showed that different SAR data were connected well in spatial and temporal domains.
Second, a model-backfeed (MBF) method was developed for optimally estimating deformation parameters. It allowed us to include a-priori knowledge and to optimize functional and stochastic models. Based upon the output from the standard InSAR deformation time series analysis method, the best deformation model of every IMP was estimated by MHT. Then, the pointwise updated functional and stochastic model for every IMP was used to repeat phase unwrapping. The method was applied for monitoring surface deformation of the Groningen gas field from 1995 to 2015 with ERS-1/2, Envisat and Radarsat-2 SAR data. The results showed that the MBF increased the number of IMPs, assigned a better deformation model for each IMP and reduced phase unwrapping errors.
Third, a systematic and optimal method for monitoring tidal flats was developed. It mainly focused on multi-polarimetric SAR data covering the Dutch Wadden Sea area. The study area was divided into multiple patches based upon their naturally spatial discontinuity. We combined an Object-Based Image Segmentation (OBIS) method with SAR images to extract waterlines and separate tidal flats and coastlines from water bodies. The method was used for mapping the Dutch Wadden Sea tidal flat dynamics from 1986 to 2020. The results are in line with LiDAR and GNSS reference observations. The study concluded that monitoring islands, sandbanks and tidal flats provided important information for the evolutionary history of the tidal flat area.
To summarize, this thesis developed a multi-sensor SAR data integration method and two time series analysis methods that were used to estimate the surface changes over Rotterdam, Groningen gas field and the Dutch Wadden Sea tidal flats. These contributions further broaden the application range of using satellite SAR imagery.