Denoising of interferometric SAR time series: towards global (slow) fault slip detection

Over the last decades, synthetic aperture radar (SAR) images and SAR interferometry (InSAR) have revolutionized Earth observation, allowing for geophysical monitoring of Earth surface processes with centimeter-to-millimeter precision. Accurate measurement of ground displacement is essential for the understanding of natural hazards, such as earthquakes. In particular, the detection of small ground (transient) displacements is of utmost importance for better imaging the dynamics of active faults, especially in tectonic settings undergoing low deformation rates. However, detecting small deformation signals in InSAR data remains a significant challenge due to the high noise level in the data (e.g., speckle noise, tropospheric and ionospheric perturbations). Multiple and successful InSAR mass processing methods, including state-of-the-art noise correction methods, have been developed over the last decade, but all rely on intensive computing of massive databases, a tedious procedure that cannot yet be applied at a global scale. Furthermore, because of the low probability of finding earthquakes in intraplate continental settings, automatic detection of such signals with InSAR data is currently out of the question, mostly due to the low signal-to-noise ratio.

Here, we develop a deep-learning-based method to denoise InSAR time series. We design a spatiotemporal attentive convolutional U-Net to retrieve small-scale deformation in noisy interferometric SAR time series, trained in a hybrid supervised and self-supervised manner on synthetic data and evaluated first on synthetic and, finally, on real InSAR time series. When applied to a time series in the North Anatolian Fault, the method effectively extracts millimeter-scale deformation associated with fault creep. The extracted deformation is consistent with independent ground truth measurements, thereby validating our method and opening the possibility of its application to diverse tectonic settings globally, as well as targeting the method to the detection of dislocation-like signals in raw SAR data, possibly optimizing the SAR interferometry processing chain, reducing the need to process entire datasets, and significantly accelerating computation.