Characterisation of strike-slip fault offsets using convolutional neural networks

The understanding of seismic recurrence relies on the chronology and magnitude of earthquake ruptures that have occurred in the past along a given fault. Knowledge of the rupture history of a fault provides valuable insights into its potential future behavior, aiding in the assessment of seismic hazard. Geomorphic evidence of faults is thus crucial for constraining models of seismic recurrence through surface rupture. The advent of remote sensing and other high-resolution datasets (such as Pleiades and SPOT satellite images) has improved tectono-geomorphological studies, promising to improve earthquake recurrence models. While manual or semi-automatic measurements of fault offsets using topographic markers like rivers have been conducted (Manighetti et al., 2015, 2020; Zielke et al., 2012), recent advances in artificial intelligence (AI) open new avenues for geoscientific applications (Ren et al., 2020) to handle the amount of high-resolution datasets.

This study takes on the challenge of measuring slip offsets of faults using a Convolutional Neural Network (CNN) applied to synthetic Digital Elevation Models (DEM). The methodology involves generating realistic synthetic landscape models (DEM) using the Landlab software (Hobley et al., 2017), simulating slip faults based on the method of Reitman et al. (2019). The approach includes creating synthetic DEMs with Landlab, incorporating fault effects such as erosion, slip rates, and variable fault zone widths. Preliminary work in this study involves automating the creation of synthetic DEMs for a 2D prototype with a variable slip fault. A regression CNN model (with three convolutional layers followed by max-pooling layers and fully connected layers) is trained on these synthetic datasets, achieving slip offsets of ±3 meters on validation data. The model is then tested on real data labeled by experts, yielding satisfactory preliminary results.

This study demonstrates the potential of CNNs for measuring slip offsets of faults using synthetic DEMs. The successful application of AI to geosciences paves the way for more efficient and automated analysis of fault activity in landscapes, thereby contributing to an enhanced assessment of seismic risks.