Geometry and Precision in Sentinel-1 Processing for Interferometric Applications

This thesis explores advanced techniques for processing Sentinel-1 Synthetic Aperture Radar (SAR) data and focuses on questions related to precision, geometry, and efficiency with applications to interferometry (InSAR). The first part of the thesis studies the coregistration and geometric modeling of Sentinel-1. First, we compare different deramping approaches for Sentinel-1 and highlight the tradeoffs between efficiency and precision. Then, we successfully apply an algorithm to fit RPC camera models from 3D-2D point correspondences on Sentinel-1 and Wordlview-3 images. The last chapter in the first part proposes a method to improve the precision of Sentinel-1 burst stitching and geometric coregistration by accounting for fine corrections of the geolocation model. This results in a time series of well-aligned, geometrically consistent Sentinel-1 mosaics. The second part of the thesis evaluates the performance of existing deep learning methods for InSAR tasks such as phase denoising, coherence estimation, and phase unwrapping. The last part presents an InSAR application on crude oil storage tanks. We show the correlation between the double difference of the InSAR phase on neighboring fixed reflectors on the tanks and the double difference of the tanks' fill ratio. Our results indicate that the reflectors on the roof of a tank move away from the satellite by around 1 cm when it fills up. Our study underlines the need for stringent requirements on the precision and processing of some localized InSAR applications. In conclusion, in this work, we highlight aspects related to the geometry and the precision of Sentinel-1 processing, which could benefit both wide-area and localized SAR and InSAR applications.