Modelling Of High-Resolution Coherent Scatterer Radar Interferometry And Methodology For Extracting Deformation

From the 1950s, as the world economy developed rapidly and the urbanization continued accelerating, the anthropic need of freshwater was significantly increased, thus leading to excessive exploitation of groundwater. The decrease of groundwater resulted in severe ground subsidence. Ground subsidence has the features of regional-scale and often occurs in urban and the accessorial suburb areas with developed economy and dense population. As a result, it causes adverse effects to economy development and human life. In China, ground subsidence mainly occurs in the Yangtze River delta, the North China Plain and the Fenwei Basin area. Ground subsidence caused huge loss to the economy of these areas, especially to that of the North China Plain. The subsidence funnels in the North China Plain impact a large area of more than 50000 km2, which is considered the largest groundwater depression all over the world. More seriously, the subsidence centers in different areas are still expanding and going to be connected as one large subsidence trough.Currently, there are two categories of subsidence monitoring methods. One is the conventional geodetic surveying (e.g., GPS and leveling) method, and another is the differential interferometric synthetic aperture radar (DInSAR) surveying method. GPS and leveling have the drawbacks of inefficiency and low spatial resolution, thus not applicable for large area (e.g., the North China Plain) ground subsidence monitoring, subsidence center detection and fountainhead exploration. DInSAR has the capabilities of all time and all weather imaging, pantoscopic and high resolution observing and high sensitivity to vertical deformations. These advantages make DInSAR be one of the most important tools for ground subsidence monitoring.However, DInSAR is affected by many negative aspects, including spatial and temporal decorrelation, atmospheric delay, orbital error, elevation error (DEM error) and other errors related to data processing. Such drawbacks significantly reduce the accuracy of DInSAR deformation estimates, and even lead to false results. In order to eliminate such drawbacks, the time series DInSAR (TS-DInSAR) methods were proposed. TS-DInSAR utilizes time series SAR images to implement coherent scatterer (CS) detection, time series differential interferometric processing, deformation modelling and estimating, and finally to separate atmospheric delay and orbital errors. The outputs of TS-DInSAR are the linear deformation rates and deformation time series. Two typical strategies for TS-DInSAR are the persistent scatterer (PS) DInSAR (PS-DInSAR or PSI) and the small baseline subset (SBAS) DInSAR (SBAS-DInSAR). For further mitigating spatial and temporal decorrelation, corner reflector (CR) DInSAR (CR-DInSAR) was proposed. CR-DInSAR utilizes CR as artificial CS for deformation modelling and estimating.In recent years, the new generation high-resolution SAR systems e.g., TerraSAR-X (TSX) and COSMO-SkyMed (CSK) were lunched for data acquisition. The availability of high-resolution SAR data promoted the application of TS-DInSAR in deformation monitoring. Besides the feature of high resolution (1 m spatial resolution by stripmap mode), these SAR systems collect images with the short wavelength X-band radar waves. Short wavelength gives them the capability of having higher sensibility to ground deformations as compared with the moderate and long wavelength SAR systems, which makes them very suitable for capturing the slow and accumulated ground subsidence in urban areas.Although the application of high-resolution SAR images brings multiple advantages for TS-DInSAR deformation detection, while the characteristics of high resolution and short wavelength result in relevant defects as well. This thesis concentrates on some of the key issues of TS-DInSAR with high-resolution SAR images, including precise co-registration of SAR images with external digital elevation model (DEM), seasonal decorrelation and preferential interferometric pair selection, using of CR in high-resolution TS-DInSAR and large gradient deformation modelling and estimating.The resolution and ground cover differences between high-resolution SAR images and the external DEM and the speckle noises of high-resolution SAR images often lead to decrease of co-registration accuracy or failiure. In this thesis, we propose a method to enhance the co-registration accuracy by combining SAR image filtering process and multilevel iterative co-registration (MICR) strategy. We first filter the SAR images to eliminate speckle noises, and then perform the MICR strategy to achieve precise matching between SAR images and DEM. The matching window size, the number of matching windows and the matching norm threshold very gradually between each implementation of the MICR strategy when estimating the offsets between SAR images and DEM. It should be noted that the gross error detection process is performed to eliminate the improper offset estimates. Moreover, each level co-registration takes the offset estimates from the last level as the initial inputs to avoid redundant operations. For testing purpose, we utilize the TSX, ASAR and PALSAR SAR images and the SRTM3 DEM as the data source for conducting the proposed co-registration method. The results demonstrate that the standard diviation of the offset polynomial fitting residules can be constrated at sub-pixel level. For further validation purpose, we extracted three sets of CS point cloud (CPC) using the TSX, ASAR and PALSAR images, respectively, and geocoded the three sets of CPC according to the co-registration parameters. The CPC fusion result shows that the three CPC sets have very good degree of overlapping. Additionally, the ground features presented by the CPC fitted well with those in Google Earth optical images. The results indicate that the proposed method is able to actualize precise co-registration between SAR images and the external DEM. This is a critical prerequisite for the subsequent TS-DInSAR data processing and analyzing.The short-wave SAR systems are more sensitive to ground target property changes that are highly related to season iterations (e.g., temperature changes). The relevant seasonal dynamics often lead to significant decorrelation in SAR interferometry, even for stable areas in SAR images. In this thesis, we propose the concept of seasonal decorrelation for SAR interferometry in stable areas (e.g., man-made ground targets) and elaborate its definition and the relevant descriptions. The TSX SAR images (40 scenes) covering Jingwu town in Tianjin are utilized for interferometric coherence estimation and seasonal decorrelation analysis. The results demonstrate that the decorrelation caused by seasonal ground target property changes is much more obvious than that related to long-time temporal (but same season) ground target changes. Further inspections indicate that the seasonal decorrelation is associated with temperature variations. In terms of the above analysis, we propose the concept of temperature baseline and elaborate its definition and the relevant descriptions, and the method of temperature baseline estimation is presented as well. Accordingly, a method for preferential interferometric pair selection based on temperature baseline threshold is proposed for identifying interferometric pairs affected by significante seasonal decorrelation. Such interferometric pairs will be removed from the interferometric pair sets, thus avoiding the negative effects by seasonal decorrelation on coherent scatterer (CS) detection. For validation purpose, the above-mentioned SAR images are used for CS detection experiment. The results demonstrate that the total number of CS is increased by more than 35% after the elimination of seasonal decorrelation interferometric pairs, while keeping those with long temporal baselines. The analysis indicates that the proposed interferometric pair selection method is capable of eliminating seasonal decorrelation effects, which is advantageous to TS-DInSAR data processing and deformation extracting.For CR application in high-resolution TS-DInSAR, we propose to explore the key issues such as CR lectotype, echo signatures and stability in high-resolution SAR images and deformation estimates accuracy. A new type detachable CR (DCR) is presented and a method for CR phase center identification is proposed. The 13 high-resolution TSX SAR images covering the Xiqing district in Tianjin are selected as data source for conducting the relevant experiments. The echo signatures (echo intensity and imagery presentations) and stability of different type of CRs (DCRs and the fixed CRs) are comprehensively analyzed, demonstrating that the echoes of DCRs are more stable than those of the fixed CRs (FCRs), and both of them are better than the natural persistent scatterers in performance of radar reflectivity. Further inspections of the SAR intensity images present the significant side lobe effect of CRs in high-resolution SAR system. Due to the side lobe effect, the adjacent image pixels around the CR phase center present high echo intensity and stability as well. This is one of the echo signatures of CR in high-resolution SAR images. For assessment of CR deformation estimates accuracy, the PSI approach is conducted for extracting deformation rates and deformation time series at all CRs and the precise leveling data is utilized for validation purpose. The numerical analysis shows that the standard deviations of the discrepancies between the two types of results are ±3.6 mm (for accumulated vertical deformations) and ±2.1 mm/yr (for vertical deformation rates), respectively, while those in the radar line of sight (LOS) direction are ±2.7 mm and ±1.6 mm/yr, respectively.For the issue of large gradient deformation modelling and estimating by high-resolution TS-DlnSAR, this thesis first proposes the concepts of spatiotemporal deformation gradient (STDG) and the spatiotemporal phase gradient (STPG) and elaborate their definitions and the relevant descriptions. The theoretical models of STDG and STPG are established. According to the proposed theoretical models, large STDG induced impacts on TS-DInSAR deformation modelling and estimating are discussed. One of the key impacts is under-estimation of deformations. For purpose of solving this problem, we propose the model confined network differencing short baseline subset (MCND-SBAS) interferometry approach and the methodology for large gradient deformation modelling and estimating on basis of the MCND-SBAS approach, the STPG updating strategy and the iterative processing chain. For testing purpose, we select Tianjin and Jingwu town (the subsidence funnel area) as the study areas and utilize 35 high-resolution TSX SAR images covering the study area for deformation modelling and estimating experiment. For comparative analysis, the TSX data are also processed by the conventional SBAS-DInSAR and the StaMPS PSI approaches. The results demonstrate that the deformations in areas with large deformation gradient are significantly under-estimated by SBAS-DInSAR and StaMPS PSI. However, the large gradient deformations (deformation rates and deformation time series) can be accurately estimated by the proposed methodology. For further accuracy assessment,147 leveling points are utilized and the two types of deformations derived by precise leveling and the proposed TS-DInSAR methodology, respectively, are compared. The standard deviations of the discrepancies between two types of accumulated deformations (deformation time series) and deformation rates (vertical direction) are ±4.7 mm and ±3.8 mm/yr, respectively, while those in the LOS direction are ±3.5 mm and ±2.9 mm/yr, respectively. The results indicate that the proposed methodology has very high level accuracy and is reliable for modelling and estimating large gradient deformations, thus can provide reliable technical support for subsidence funnel monitoring and subsidence center identification.On the basis of the above validated deformation (vertical direction) results, the subsidence in the study area is analyzed. Multiple subsidence funnels are detected in this area, forming large subsiding troughs (see Fig.7-41). The maximum subsidence is -190 mm (the negative sign presents moving away from the TSX satellite, i.e., subsidence) in about 1 year. More than half (around two-thirds) of the area show subsidence magnitudes over -50 mm. Further analysis indicates that the subsidence distribution pattern is highly related to the land use categories (e.g., industrial parks, urban residential quarters, towns and villages) of this area, which can provide scientific references for preventing and controlling subsidence in the study area.