Get Complete Project Material File(s) Now! »

Overcoming Limitations due to Decorrelation Noise

The reliability and accuracy of an interferogram obtained by SAR processing is limited by the problem of decorrelation on interferometric coherence (i.e., similarity between the acquisitions). The decorrelation can be temporal decorrelation which arises from changes on surface scatterer properties due to long period cover (such as vegetation growth, soil moisture, glacier motion, etc.) and increase with time between master and slave acquisitions. It can also be spatial decorrelation due to long baselines (distance between two antennas). Quantitative coherence information is very important for the performance of SAR processing. This decorrelation limitation arising from different sources (as well as other limitations due to random atmospheric delays for example) can be overcome by using new state-of-the-art InSAR time series methods combining information from multiple SAR acquisitions instead of just two images. Currently, there are two families of time-series InSAR technique incorporating information from multiple SAR acquisitions – Persistent Scatterer (PS) (e.g. Ferretti et al., 2001, Perissin and Ferretti, 2007, Hooper et al., 2007) and Small Baseline Subset (SBAS) (e.g. Berardino et al., 2002, Lanari et al., 2004, Biggs et al., 2007, Hooper, 2008) methods or a combination of both (Hooper, 2008) that identify areas where ground surface is least affected by spatial and temporal decorrelation.

Persistent Scatterer Method (PS)

As discussed above, the PS approach is based on the use of co-registered long series stack of SAR images (the larger the number of SAR data, the more precise and robust the results are) to generate differential interferograms with respect to one common master. It exploits the radar return signal backscattered from phase stable targets that function as corner reflectors due to their coherent phase center. It is developed to overcome some of the limitations of conventional InSAR decorrelation problems that give rise to disconnected areas in the interferograms in space and time (Hooper, 2006) and makes the geodetic interpretation difficult. Unlike the Small Baseline method, the single-master interferogram method uses all possible interferometric combinations regardless of their spatial and temporal baselines values and temporal differences by optimizing the each resolution call containing single point scatterer. On the other hand, small baseline method is another InSAR stacking method that relies on appropriate combinations of SAR image pairs characterized by a small orbital separation (baseline) and with no selection of specific scatterers. PS-InSAR method is able to generate the average velocity and time series of displacements by measuring the displacements obtained with the set of interferograms.
As of today, there are various methods for identifying the PS pixels in interferograms (e.g. Ferretti et al. 2001; Crosetto et al., 2003, Adam et al., 2005 etc.). In this dissertation, we used the Permanent Scatterers Interferometry Synthetic Aperture Radar (PS-InSAR) technique to compute the displacement time series of each stable Permanent Scatterer pixel (PS) in the study area, based on a single master network in the Stanford Method for Persistent Scatterers (StaMPS) (Hooper, 2008; Hooper et al., 2012) package, that takes advantage of the spatial correlation between pixels. StaMPS uses only the selected pixels that show stable phase and amplitude noise characteristic in time to compute average velocities and time series. This method is considered to be a modified version of PS algorithm and is applicable in areas undergoing non-steady deformation, such as aseismic slip, without using prior knowledge of the temporal deformation model.
Figure 3.1 : Schematic diagram illustrating the difference of the phase simulation for 100 acquisition iterations for (a) a pixel with distributed scatterers showing dispersed signal values, and (b) a pixel with a dominant Persistent Scatterer pixel with small phase variations (modified from Hooper et al. (2007)).
The cartoons in Figure 3.1 represent the scatterers contributing to the phase of one pixel in an image. In case of non-stationary resolution elements such as vegetated or cultivated areas where scatterers “move” with respect to each other between satellite passes, the phase sum changes (Figure 3.1-a). In contrast, if the scatterers are dominated by a stable large reflector called a persistent scatterer (can be either manmade or natural features such as buildings, rock outcrop, towers etc.), it results in a stable phase over time (Figure 3.1-b).
I am giving in the next sections a detailed description of the procedures used in each step of the interferogram formation and the StaMPS PSI approach employed in this dissertation.

Procedures: from SLC to geocoded interferograms

For processing interferograms, we used the open-source radar interferometry software, Generic Mapping Tools Synthetic Aperture Radar (GMTSAR) developed by Sandwell et al. (2011a,b) that is based on Generic Mapping Tools (GMT; Wessel et al., 1998). Forming an interferogram using two Single Look Complex (SLC) images involves three basic steps using GMTSAR’s main processors: 1) a preprocessor to convert each satellite data type and orbital information into a generic format; 2) an InSAR processor to focus, co-register images, remove topographic phase, and form the complex interferogram; 3) post processor based on GMT, to filter the interferogram and construct interferometric phase, amplitude, coherence and phase gradient. Unlike many other InSAR systems such as ROI_PAC, GAMMA and DORIS, GMTSAR relies on sub meter orbital accuracy, which greatly simplifies the SAR processing algorithms.
The focused SAR image contains backscatter amplitude (intensity) and phase (range) information for each pixel. This is the complex image. The amplitude component is the strength of backscattered radar signal from each resolution element that contains uncertainties that stem from terrain scattering features such as surface roughness, soil moisture and dielectric constant. The phase is the fraction of the full wave that represents the precise distance between satellite and target on the ground. Multiplying the first image by the complex conjugate of the second image forms the complex interferogram. The first step is coregistration of slave and master images in sub-pixel accuracy. In GMTSAR, this is done by 2 dimensional cross correlation method of multiple small sub-patches taken from slave and master acquisitions (xcorr.c in GMTSAR). Then GMTSAR uses 6-parameter (3 along range and 3 along azimuth) affine transformation to align the images using the robust 2-D fit to the offset data (esarp.c in GMTSAR). After this step, the phasediff.c program computes the phase difference of the images by removing the effect of curved earth and baseline dependent topographic phase contribution prior to cross multiplication. The output file contains the real and imaginary components of a complex interferogram. After all the position and phase corrections are applied, a spatial low-pass filter (filter/snaphu) is applied in order to enhance the results of real and imaginary components of the interferogram and compute standard products of amplitude, phase and coherence. GMTSAR’s unwrapping processor is based on snaphu (snaphu.c) unwrapping algorithm. Detailed description method of SNAPHU unwrapping method can be found in Hooper (2010).
The final step in the processing is to geocode all of the final products by transforming from radar (range and azimuth) coordinate system to geographical coordinate system (longitude and latitude).

READ  Inpatient blood glucose management of diabetic patients in a large secondary hospital

Procedures: from Interferograms to PS time-series

The idetification of the PS pixels for the time series analysis is performed using single-look coherent interferograms. Following interferogram processing, we use the Permanent Scatterers Interferometry Synthetic Aperture Radar technique to compute the displacement time series at each PS in the study area using a single master network with StaMPS (Hooper, 2008; Hooper et al., 2012). StaMPS is a Matlab-based open source software that has been developed partially at Stanford University, the University of Iceland and Delft University of Technology. This PS-InSAR technique is an advanced method used to monitor surface deformation caused by tectonic or anthropogenic activities. The PSI method allows monitoring subtle and slow deformation signal, mitigating errors due to atmosphere, digital elevation model and orbital inaccuracies using spatial and temporal filtering. Unlike the amplitude based approach proposed by Ferretti et. al. (2001), StaMPS does not assume any priori hypothetical model of the deformation. StaMPS uses both amplitude and phase information to identify PS points. As the first step of pre-processing, the input dataset is converted into a format readable by StaMPS and an area of interest is chosen, that must be included in all slave images. In an initial step, the selection of PS points is performed based on their noise characteristics, using an amplitude dispersion analysis. The amplitude dispersion criterion is defined by: DAmp = (σAmp⁄mAmp) where σAmp and mAmp are the standard and mean deviation of the amplitude in time, respectively (Ferretti et al., 2001). In the present studies, we selected a threshold value of 0.42 for DAmp that minimizes the random amplitude variability and eliminates highly decorrelated pixels in some areas covered with vegetation, agricultural fields or snow. Once coherent PS points have been selected based on this amplitude analysis, the PS selection is refined by phase analysis in a series of iterations, which allows the detection of stable pixels even with low amplitude. The next steps of the StaMPS processing includes correction of the residual topographic component and 3D unwrapping of the PS phase both in time and space.
Unwrapped interferograms after PS selection are corrected from stratified tropospheric delays using the freely available Toolbox for Reducing Atmospheric InSAR Noise (TRAIN, Bekaert et al., 2015), based on the ERA-Interim (ERA-I) global atmospheric model. Corrected interferograms are then used to compute mean velocity fields in StaMPSv maps in study areas.


Earthquakes are the phenomena that caused by rapid release of elastic strain energy by a sudden slip events propagate along a fault. Estimation of the energy released by elastic strain for the future possible events are complicated by the occurrence of asesismic creep. A better understanding of seismic and aseismic slip behaviors of active faults is a crucial challenge for reliable assessment of seismic hazard (Avouac, 2015) as aseismically slipping part of the fault reduces the seismic moment that capable to produce earthquake. While most active faults are interseismically locked within the resolution of geodetic measurements and release abruptly the strain that has accumulated over long time, some other faults in particular locations slip continuously or transiently with an absence of destructive earthquakes (e.g. Thatcher, 1979; Burford & Harsh, 1980). Aseismic slip occurring within or around the seismogenic zone along continental faults is usually referred to as creep and it is an important indicator of the shear stress. Determining the lateral variation and temporal characteristic (steady-state or transient) of aseismic creep is highly important for understanding the mechanism of fault behavior (Galehouse, 1997; Lienkaemper et al., 1997, 2001). In this regard, knowing how slip and creep rates vary spatiotemporally has profound implications for seismic hazard assessment (Carpenter et al., 2011).
Studies based on Interferometric Synthetic-Aperture Radar (InSAR) and Global Positioning System (GPS) satellite observations until 2012 have shown that the central segment of the August 17, 1999 Mw 7.4 Izmit earthquake on the North Anatolian Fault began slipping aseismically following the event. In the present study, we combine new InSAR time series, based on TerraSAR-X and Sentinel 1A/B radar images acquired over the period 2011-2017, with near field GPS measurement campaigns performed every 6 months from 2014 to 2016.

GPS Data and Processing

To complement InSAR Sentinel-1 data in modelling the most recent creep rate along the Izmit section of the North Anatolian Fault, we used GPS data that we acquired from 2014 to 2016 during five campaigns (one every six months) along near field zone. Figure 2 shows the corresponding GPS velocity field for the study area. Details on GPS sites, velocities and associated uncertainties are provided in Appendix A. GPS data were processed using the GAMIT/GLOBK (V10.6) software developed by MIT (King and Bock, 2004; Herring et al. 2015), following the processing strategy in Ergintav et al. (2009).

Table of contents :

Thesis Roadmap
2.1 The Earthquake Cycle
2.1.1 Interseismic period
2.2.1 Aseismic creep and seismic potential
2.3 The North Anatolian Fault (NAF)
2.3.1 Postseismic creep along the North Anatolian Fault
2.4 Ground Subsidence
3.1 Introduction
3.2 Principles of InSAR
3.2.1 Contributors to signal phase
3.3 Overcoming Limitations due to Decorrelation Noise
3.4 Persistent Scatterer Method (PS)
3.4.1 Procedures: from SLC to geocoded interferograms
3.4.2 Procedures: from Interferograms to PS time-series
4.1 Introduction
4.2 GPS Data and Processing
4.3 InSAR Data and Methodology
4.3.1 Multi-sensor insar time-series
4.3.2 PS-InSAR Processing
4.4 Results
4.4.1 Mean Line-of-Sight Velocity Fields
4.4.2 Fault-parallel horizontal velocities versus vertical velocities
4.4.3 Horizontal Creep Rate Distribution Along Fault
4.4.4 Vertical Motion Along the 1999 Rupture
4.5 Modelling Profile Velocities
4.6 Temporal Evolution of Creep
4.7 Discussion
4.7.1 Spatial Distribution of Creep and Geometrical and Lithological Control
4.7.2 High Temporal Resolution InSAR Data Reveal Burst-Like Behavior of Creep
4.8 Conclusion
5.1 Introduction
5.2 Background of Study Areas
5.2.1 Geological setting
5.2.2 Study Area 1: Haramidere valley and Avcilar neighborhood
5.2.3 Study Area 2: Ayamama river valley
5.3 Datasets
5.4 Results and Discussion
5.4.1 Main deformation patterns
5.4.2 Site 1 : Haramidere valley
5.4.3 Site 2 : Ayamama river
5.4.4 Sites 4, 6 and 3
5.4.5 Site 5
5.5 Conclusion
6.1 Introduction
6.2 Study Area
6.3 Datasets
6.4 Deformation Decomposition of Sentinel -1 Data
6.5 Results
6.5.1 InSAR-derived land subsidence maps
6.6 Discussion
6.6.1 Impact of the looking geometry on self-consistency checking between the ascending and descending InSAR observations
6.6.2 Water table variations and InSAR time series
6.6.4 Compressibility of the aquifer in the Bursa plain
6.6.3 Lithological and tectonic control over land subsidence
6.7 Conclusion


Related Posts