Characterising Bias and Noise in Interferometric Synthetic Aperture Radar Time Series

Natural and anthropogenic hazards associated with the deformation of the Earth’s surface pose substantial risks to societal health and safety. One of the most effective methods for categorising and monitoring these hazards is through the measurement of deformation over time. Although various methods, such as levelling surveys and GPS monitoring, can be employed to achieve this, only Interferometric Synthetic Aperture Radar (InSAR) utilising space-borne (or aerial) platforms like Sentinel-1 is capable of monitoring the majority of the Earth’s solid surface on a regular basis.

The substantial volume of SAR data generated by missions like ESA’s Sentinel-1 satellite enables near-real-time deformation monitoring through time series InSAR. While this approach has proven highly successful in the field of geodesy, several challenges persist. This thesis addresses key challenges, primarily focusing on phase noise estimation, coherence reliability, change detection, and phase bias effects.

Reliable deformation measurements depend on accurately estimating the phase noise of pixels in InSAR time series data. Coherence estimation can be improved by using groups of pixels with similar scattering characteristics, referred to as sibling ensembles. Many methods use amplitude similarity for sibling selection, raising questions about whether incorporating phase would improve coherence estimates. I test and optimise a method known as Similar Time series Interferometric Phase, or STIP, which integrates phase information into ensemble selection.

To ensure sibling ensembles continue sharing the same scattering characteristics and produce accurate coherence is important for long time series. I use the shared scattering characteristics between nearby pixels to develop a novel change detection methodology with a focus on high-resolution InSAR data.

Finally, I investigate phase bias, which affects short-term multilooked interferograms and can introduce deformation measurement errors. By analysing closure phase, a key marker of phase bias, for both C-band and L-band InSAR across different land cover types, I gain insights into the underlying scattering mechanisms. My findings show that the observed, consistently positive closure phase bias can be successfully reproduced in a simulation framework by using a temporally inconsistent and asymmetric phase noise signal. I also characterise how the mean closure phase increases with the number of looks before converging an a stable plateau, and develop a mathematical model to describe this convergence. This behaviour is understood to be a consequence of the process of multilooking suppressing the interference cross-terms between scatterers, which in turn reveals the underlying systematic bias.

This thesis presents methods and insights that enhance the accuracy and reliability of time series InSAR, particularly for long-term deformation monitoring and change detection. This leads to improved usefulness of deformation monitoring for the health and safety of society, both in relation to natural and anthropogenic hazards.