Efficient High-Precision Time Series Analysis for Synthetic Aperture Radar Interferometry

Interferometric Synthetic Aperture Radar (InSAR) is a modern geodetic technique. Exploiting differential phase measurements, it retrieves Earth surface topography and deformation with an unprecedented spatial coverage. The precision of the technique varies between centimeter to millimeter level, dependent on the calibration of the error sources. Among such sources is signal decorrelation which compromises the precision of interferometric phase. Various time series analysis techniques have been developed to tackle the decorrelation problem. They enhance the phase precision albeit at high computational complexity. The latter has hindered the applicability of these techniques in processing long time series. Over the last few years the new generation wide-swath SAR missions have opened unique opportunities for rapid global deformation monitoring. However, the demanding processing of the unprecedented data volume from these missions poses a challenge to the conventional interferometric techniques. This dissertation responds to this challenge by proposing novel algorithms for efficient phase, and consequently deformation, estimation. Special attention is paid to the achievable estimation performance. The following summarizes the milestones in fulfilling these objectives. The performance of the state-of-the-art phase estimators is assessed and compared with respect to the Cramer-Rao Lower Bound (CRLB). This evaluation highlights a trade-off between computational efficiency and estimation precision in time series analysis. The optimization of this trade-off is the main objective behind the proposal of two new phase estimators. Eigen-decomposition-based Maximum-likelihood-estimator of Interferometric phase (EMI) is the first proposal. In its design, the model complexity of conventional phase estimators is increased to improve the estimation performance. The inversion of the generalized model is sought via Maximum Likelihood Estimation (MLE). In the interest of gaining computational efficiency, the reference MLE is approximated and reformulated into a Lagrangian. An arsenal of highly optimized numerical recipes is available for the efficient solution of the latter. However, similar to the state-of-the-art techniques, EMI requires the reprocessing of the entire time series for inclusion of each new ingested acquisition in its framework. The second proposed estimator accommodates stream processing of the ingested acquisitions, thereby provides high operational flexibility for systematic deformation monitoring. This Sequential Estimator efficiently exploits the abundant data in the time series via data compression and generation of high SNR artificial interferograms between the compressed and ingested data content. The artificial interferograms are shown to assist the Sequential Estimator in maintaining performance close to computationally demanding estimators. The proximity of performance to CRLB, elimination of redundant data processing, reduction of the required data storage and working memory capacity are among the efficiency criteria for the design of the Sequential Estimator. The combination of the Sequential Estimator and EMI is the ultimate suggestion of this dissertation for Big Data analysis. This combined approach is demonstrated in wide area processing. Its performance in terms of precision and accuracy of phase estimation is thoroughly investigated. Such error analysis assists in the quantification of the achievable InSAR performance in deformation estimation. Finally, the performance of InSAR in 3D deformation estimation is assessed. Optimized acquisition geometry is proposed for achieving high-precision 3D deformation monitoring