| Atmospheric effects are one of the limiting error sources in repeat-pass SAR interferometry measurements. The effects have however not been fully characterized and understood, and the existing methods for mitigating the effects are largely at a very preliminary stage and far from optimal. This thesis mainly aims to systematically characterize the effects, and to develop algorithms for modelling and correcting the effects.; In characterizing atmospheric effects on InSAR, the isotropy/anisotropy property of the atmospheric signals in SAR interferograms is first studied with Radon transform. The results reveal that the signals in all the interferograms studied are anisotropic although the anisotropic patterns are different between the interferograms. The Gaussianity/non-Gaussianity of the atmospheric signals is then examined with the Hinich non-Gaussianity test. This is the first time that the property is studied and the results show that the signals are non-Gaussian, contradicting to the common belief of Gaussian distribution. The spectral characteristics of the atmospheric signals are finally studied with the method of bispectrum decomposition for the first time. The bispectrum-reconstructed power spectra exclude the contributions from radar normal noises, and therefore better reflect the characteristics of atmospheric signals than conventional power spectra. Although the conventional and bispectrum-reconstructed power spectra differ in details, they on the whole both follow the power law commonly associated with the Kolmogorov turbulence. The study provides improved understanding of various properties of the atmospheric effects that is important for designing algorithms to mitigate the effects.; In modeling and correcting atmospheric effects on InSAR with ground-based data, we propose to integrate continuous GPS and surface meteorological observations for improved resolution. The Ordinary Cokriging (OCK) algorithms are suggested to integrate the two types of data. The OCK paradigm is then modified into a Simple Cokriging with varying local means (SCKlm) algorithm to better incorporate the elevation-dependent nature of the atmospheric delays. Cross validation tests show that the proposed method can correct 37% atmospheric errors for one-day intervals and 39% atmospheric errors for ten-day intervals. Besides, the proposed model reduces to Delacourt's model under poorest conditions, and is easily expanded when more ground-based data sources are available. (Abstract shortened by UMI.)... |
