Detecting creep transients in InSAR timeseries using deep neural networks

Yuexin Li, Roland Bürgmann, & Gareth J. Funning

Submitted September 11, 2022, SCEC Contribution #12414, 2022 SCEC Annual Meeting Poster #190

Creeping faults in California exhibit both steady and episodic slow slip behaviors. Growing evidence indicates that shallow fault creep can be triggered by nearby or remote earthquakes, modulated by elevated pore pressure in the fault, and composed of aseismic slip transients in addition to background steady creep. Interferometric Synthetic Aperture Radar (InSAR) measures high-resolution large-scale ground deformation on a routine basis. The short (6-12 days) recurrence interval of the Sentinel-1 satellite makes it possible to monitor short-term creep transients. Theoretically, InSAR timeseries can reach sub-centimeter accuracy if the network of interferograms is properly designed and atmospheric noise contributions are minimized. However, the detectability of low-amplitude transient signals is still largely limited by the contamination of atmospheric noise in SAR acquisitions.

Traditional atmospheric correction methods include applying an empirical model relating topography and atmospheric delay (e.g. linear, quadratic), using an external auxiliary dataset to estimate the atmospheric delay (e.g. GNSS, ERA5, MODIS, GACOS), and applying temporal filtering methods. Even though these methods have proven effective for estimating long-term interseismic rates, it continues to be challenging to extract low-amplitude transient signals from InSAR timeseries. In this study, we develop a neural network structure that is designed for detecting shallow creep transients, based on the encoder-decoder structure proposed by Rouet-Leduc et al. (2021). Specifically, we evaluate a series of neural network structures trained with >10^7 synthetic timeseries incorporating synthetic and real noise features and apply the denoising method to creeping faults in California to identify transient creep signals.

Li, Y., Bürgmann, R., & Funning, G. J. (2022, 09). Detecting creep transients in InSAR timeseries using deep neural networks. Poster Presentation at 2022 SCEC Annual Meeting.

Related Projects & Working Groups
Stress and Deformation Over Time (SDOT)