Automatic detection of slow slip events using InSAR data: Application to the North Anatolian Fault

A better understanding of aseismic slip dynamics throughout the seismic cycle is essential to refine seismic hazard estimates. Analysis of the Interferometry Synthetic Aperture Radar (InSAR) time series in the last decades has proved its efficiency to detect and characterize slow slip events (SSE), especially on strike-slip segments. However, the implementation of automatic SSE detection methods is needed to overcome the large incoming flow of data. Here, we adapted the geodetic matched filter approach developed for GNSS time series by Rousset et al. (2017) to InSAR time series. The method is computing physics-based dislocation slip models corresponding to synthetic reconstructions of SSEs, that are correlated with InSAR time series, taking advantage of the high spatial density of InSAR observations. By comparing true and false detections on synthetic tests including InSAR realistic noise, we derive probabilistic estimates of the true detections as a function of SSE magnitudes and depths. We show that this method enables the detection with ≥ 90 % confidence of shallow SSEs with magnitudes larger than 4.5 using horizontal east-west InSAR time series. And it can detect events with magnitude larger than 4.25 with ≥ 45 % confidence. We applied this method along both creeping segments of the North Anatolian Fault - Izmit and Ismetpasa, by using the InSAR time series from 2016 to 2021 automatically processed in the framework of the FLATSIM project between CNES and FormaTerre by using Sentinel-1 SAR images and based on the NSBAS processing chain (Doin et al., 2011; Thollard et al., 2021). It detected without any prior knowledge three transient events already reported by previous studies along the Izmit segment (Aslan et al., 2019; Neyrinck et al., 2024), and two transient events also already reported by previous studies along the Ismetpasa one (Jolivet et al., 2023; Özdemir et al., 2025). Based on a weighted stacked time series associated with the detections, we estimate a magnitude for these events ranging from 4.0 to 5.0, also compatible with previous estimates. Applying this method on worldwide strike-slip fault segments may allow a rapid detection and a first order characterization of transient slip events.