Using Deep Learning to harness InSAR data for creep detection

Abstract:

Fault creep is one of the main mechanisms that accommodates slip along faults. Because of the lack of seismic signals, detecting fault creep, either secular or episodic, remains a challenge, especially in remote places. Here we explored deep learning methods to automatically detect fault creep in InSAR data. Because of the limited availability of real data with fault creep signal, we generated synthetic data to train neural networks. We trained a generative adversarial network (GAN) and a convolutional neural network (CNN) to automatically de-noise and classify images. Real data from the Salton Sea was used to test the performance. We found that the method can reliably detect fault creep with displacement over 1 cm. The reliability of the detection decreases with the signal to noise ratio. The performance can be improved by training on both real and synthetic data. Our results suggest that this method can be used routinely and will provide better constraints on fault creep.