A Robust Methodology For Inter-comparison Of Sentinel-1 InSAR Products

Abstract:

Sentinel-1A & -1B offer a six-day revisit cycle and unprecedented coverage of Europe, with freely available data. This capability of the Sentinel-1 mission addresses limitations of cost and data availability, and provides research and commercial opportunities e.g. a European ground motion map. A European Ground Motion Service (EU-GMS) is currently under development, sponsored by the European Environment Agency, to provide consistent and reliable information on ground motion over Europe and across national borders, with millimetre accuracy. The ground motion results will be derived from time series analyses of Sentinel-1 data, most likely using different Persistent Scatterer (PS) and Distributed Scatterer (DS) InSAR approaches. To make the outputs useful for operational applications, quality assessment of ground motion maps is a fundamental priority, and an important aspect of quality assessment is data consistency. The nationwide/international ground motion map will be likely processed by multiple suppliers and their products can differ in terms of different metrics, such as density and coverage of measurement points, estimated deformation rate, and time series. Therefore, there is a need to assess and ensure consistency of InSAR results.

Our main goal in this research is to develop and test a fair and robust methodology capable of assessing the similarities and differences between results from different InSAR processing chains, and to recommend a validation strategy for any nationwide/international (e.g. UK/EU) ground motion map. We base our approach on the Terrafirma Validation Project (EU/ESA Global Monitoring for Environment and Security (GMES) programme) (Crosetto et al. 2008), but tackle its limitations as follows: 1) As end-users require geocoded InSAR data, we compare all the datasets in geographic rather than radar coordinates. 2) We avoid assuming that any reference InSAR processing is the “truth”. 3) We define several polygons with different land cover types and stability. 4) We do not limit the time series processing to PSI algorithms and are open to any other methodologies e.g. both PS and DS InSAR processing. 5) We work with Sentinel-1 imagery.

Our approach can be split into pre-processing and inter-comparison stages. The pre-processing stage includes checking global consistency between the InSAR datasets, re-referencing in the time and space domains, making an identical grid and defining different polygons. Then, the deformation velocities and time series, density and coverage of measurement pixels are compared and analysed by extracting some meaningful statistics. We use InSAR results from the Clyde Gateway of the Glasgow City Region in UK to test our methods. This is an area of particular interest to the Natural Environment Research Council (NERC) as it is the British Geological Survey (BGS) geothermal energy research field test site of the UKGEOS project (Bateson and Novellino 2019). We have access to multiple Sentinel-1 InSAR data products for this area, including data from SatSense, processed using a modified RapidSAR algorithm (Spaans and Hooper 2016), from TRE-ALTAMIRA, processed using the SqueeSAR algorithm (Ferretti et al. 2011), and from GAMMA-IPTA, processed using PSI at BGS. We used these datasets as well as our own analysis of Sentinel-1 using the Stanford Method for Persistent Scatterers (StaMPS) algorithm (Hooper et al. 2007). The results show that all the InSAR datasets detect similar deformation signals in the deforming polygon with all velocities consistent at the 1.1 mm/yr level (1 sigma). However, the InSAR products are not completely identical.

One of the most striking differences between different InSAR methods is density and coverage of selected pixels. In general, the results of our comparison show that those methods that take advantage of both PS and DS, and benefit from making all possible interferograms, are more successful at extracting the maximum information (density and/or coverage) from the SAR stack. However, due to the short baseline of the Sentinel-1 interferograms, some DS pixels can remain coherent in a single-master interferogram network and would be identified as PS pixels in some PS InSAR processing methods. In addition to considering both PS and DS, other factors such as the temporal sampling of signal, the configuration of the interferometric network, whether oversampling of the original images is applied, and the specific thresholds imposed on signal-to-noise ratio (SNR) for pixel selection, can all have a major impact on the density of measurements. There are also some systematic effects in difference maps between different InSAR products, which are mainly due to different approaches to dealing with long wavelength trends and atmospheric phase screens (APS). Different precise geocoded coordinates for the common selected pixels is another discrepancy between the InSAR datasets. Some qualitative indicators including spatial resolution, frequency of update and latency period are the source of inconsistencies between the InSAR providers. We discuss the reasons for these differences and make some recommendations for any future nationwide/international InSAR product based on our comparison results. Any future national or international ground motion service using Sentinel-1 InSAR will need to instigate a validation process to ensure data meet minimum standards and are consistent across borders. We propose some requirements for the validation activities.

References:

Bateson, L., & Novellino, A. (2019). Open Report: Glasgow Geothermal Energy Research Field Site - Ground motion survey report British Geological Survey, Available:http://nora.nerc.ac.uk/id/eprint/524555/1/OR18054.pdf

Crosetto, M., Monserrat, O., & Agudo, M. (2008b). Validation of existing processing chains in Terrafirma stage 2: Process analysis Report-Part 2: IG inter-comparison. ESA GMES Service Element, Institut de Geomatica.

Ferretti, A., Fumagalli, A., Novali, F., Prati, C., Rocca, F., & Rucci, A. (2011). A New Algorithm for Processing Interferometric Data-Stacks: SqueeSAR. IEEE Transactions on Geoscience and Remote Sensing, 49, 9,3460-3470, doi:10.1109/TGRS.2011.2124465.

Hooper, A., Segall, P., & Zebker, H. (2007). Persistent scatterer interferometric synthetic aperture radar for crustal deformation analysis, with application to Volcán Alcedo, Galápagos. Journal of Geophysical Research: Solid Earth, 112, B7,doi:10.1029/2006JB004763.

Spaans, K., & Hooper, A. (2016). InSAR processing for volcano monitoring and other near-real time applications. Journal of Geophysical Research: Solid Earth, 121, 4,2947-2960,doi:https://doi.org/10.1002/2015JB012752.