Fig. 3: Application to real data: the North Anatolian Fault 2013 slow earthquake.

In order to identify ground deformation signals in the noisy COSMO-SkyMed InSAR time series, we create a deep autoencoder that has an input size equal to the size of each frame of the time series, 200 × 650 pixels, and the same parameters as the autoencoder trained on synthetic data, shown in Fig. 1. Inputs are the InSAR time series and the topography of the same area (not shown). The autoencoder outputs ground deformation (bottom plot). The ground deformation is manifest as an offset across the fault. The deep autoencoder finds a strong slip signal of about 1 cm (in LOS) on the fault, in agreement with previous expert analysis of the time series¹⁴, with no a priori knowledge of the fault’s existence. a. Seismic setting of the region of the creeping section of the North Anatolian Fault. Thick red lines are the main faults of the NAF system, separating the Eurasia plate from the Anatolia microplate. Thin red lines are other mapped structures. Colored lines indicate the extent of historical ruptures. b Input raw time series from COSMO-SkyMed data (a subset of the data from Rousset et al., 2016). Color is the apparent range change between the ground and the satellite. c Denoised cumulative deformation as output by the deep autoencoder. The color scale shows ground deformation in the direction of the LOS. Dark lines are the surface trace of the NAF, shown here for reference. Thin dashed lines indicate the cross-sections shown in Fig. 4.