Multi-Mode SAR Interferometry for High-Precision DEM Generation

Abstract

SAR Interferometry (InSAR) is a well established remote sensing technique widely employed for the retrieval of topographic information. The relative vertical accuracy of a Digital Elevation Model (DEM) generated through InSAR techniques depends, among other factors, on the spatial separation between master and slave sensors and on the signal correlation between both datasets. In monostatic single-pass configurations, master and slave data are acquired simultaneously by sensors mounted on the same platform. As a consequence, the single-pass interferogram does not present temporal decorrelation and is less affected by spatially correlated artifacts. However, the fixed baseline limits the achievable relative vertical accuracy. On the other hand, repeat-pass configurations offer a flexible choice of baseline enabling the retrieval of very fine elevation measurements, provided that the temporal decorrelation can be kept small. Nevertheless, since the datasets are acquired independently, the interferograms are subject to artifacts due to, e.g., aircraft motion or propagation in the inhomogeneous atmosphere. Hence, the extraction of precise terrain information from repeat-pass data requires an accurate phase calibration, i.e., the removal of all undesired phase biases. In this thesis, the joint use of single- and repeat-pass datasets in a multi-mode configuration is proposed in order to profit from the stability of the single-pass derived DEMs in relation to spatially correlated artifacts, as well as from the robustness to noise associated with large baseline acquisitions. As the large-baseline interferometric phase is highly sensitive to the unknown topography and to systems errors and other artifacts, large phase variations between neighboring pixels are observed. The former in association with elevated noise can compromise the phase unwrapping, thus preventing the retrieval of accurate height measurements. In order to overcome this limitation and to promote a robust phase calibration, the use of data simultaneously acquired with an additional carrier frequency is proposed. Hence, the final DEM is extracted from a Dual-Frequency and Dual-Baseline (DFDB) dataset with the help of new algorithms developed for the proposed configuration. Specifically, a method for the mitigation of artifacts in the single-pass data due to multiple reflections in the aircraft is suggested. Furthermore, a dual-channel region-growing algorithm is proposed for efficient phase unwrapping. An active-contour-based unwrapping error correction strategy is developed for the treatment of residual errors associated with challenging terrain characteristics. A DFDB framework is proposed for the calibration of residual constant and linear baseline errors and global phase offsets, including alternatives in the complex domain and considering the use of external reference data. Finally, the joint estimation of the underlying topographic information is discussed and the problem of the correction of spatially correlated artifacts in the repeat-pass data and noise reduction are tackled. All developed algorithms are validated with DFDB data acquired with the F-SAR sensor during two campaigns in Germany: the first over tidal flats in the Jade Bight, North Sea; and the second over a calibration test-site in Kaufbeuren. In both cases, the proposed methodology allows the generation of elevation models with relative and absolute vertical accuracies in the order of centimeters. Lastly, the developed phase unwrapping framework is applied to TanDEM-X large-baseline data, demonstrating the use of the algorithm for single-pass spaceborne datasets with multiple acquisitions.