Bayesian Inference, Applications in Persistent Scatterer Interferometric Synthetic Aperture Radar

Coherent Synthetic Aperture Radar (SAR) imagery and its potential in interferometric analysis widened the horizon of earth observation and geodetic measurements in the past decades. Overcoming the error sources in conventional Interferometric SAR (InSAR) techniques, the advanced methods of Persistent Scatterer Interferometry (PSI) and differential SAR Tomography (TomoSAR) have been able to retrieve the geophysical signals of interest from SAR imagery. The key concept in such techniques is constraining the analysis to long time-coherent scatterers available in the imaged scene. Having the main role in such techniques, the coherence of the scatterers must be assessed prior to the interferometric analysis. In this regard, the amplitude time series can be exploited not only to assess the long time-coherency of the scatterers but also to investigate partial stability in a limited time interval of the acquisition period, the latter resulting in detection of temporal coherent scatterers. Exploitation of such temporal scatterers in advanced InSAR methods enhances the information content of the resulting products. The focus of this thesis is on the investigation of amplitude time series of interferometric stacks in order to detect long as well as partial time-coherent scatterers as the target points for the advanced interferometric methods. In the framework of this thesis, a generic Bayesian inference software is developed in order to deal with the detection, estimation and model selection involved in the time series analysis. The inference is defined based on the Bayesian network which summarizes the probabilistic dependencies as well as stochastic characteristics of the random variables of a defined model. The formation of such a probabilistic network together with exploitation of numerical approaches in the inversion renders the inference framework highly capable in dealing with any arbitrary inverse problem. The developed software is thus highly adaptable to any InSAR related problem and is capable of handling the non-linearity of the corresponding models. Using the developed Bayesian inference software, the time-coherence is investigated via estimation of the Signal to Clutter Ratio (SCR) of the scatterers in each resolution cell. The relation between the estimated SCR and phase coherence is assessed through simulations and application to TerraSAR-X data. Based on this result an algorithm is proposed for detection of the temporal coherent scatterers and estimation of the time interval in which the coherence occurs. The proposed algorithm is also evaluated via simulations as well as application to TerraSAR-X data.