Detection of faults from 2D seismic data using multi-attribute analysis and artificial neural network: a case study from Nekor Basin, North Morocco

International audience Al Hoceima region, located in northern Morocco, is frequently stroked by earthquakes associated with tectonic faults in and around the Nekor Basin. In particular, the Al Hoceima area is crossed by a network of active faults, indicating the undergoing of this region under active deformation associated with the convergence between Africa and Eurasia. Here, we attempt to highlight a set of faults from 2D seismic data acquired over the Nekor Basin using artificial neural networks (ANNs) com-bined with seismic attributes. We condition the seismic data to remove random noise and expose the geological structures, and then we use it to define a group of attributes sensitive to the presence of faults. The selected attributes were trained through a fully connected multi-layer perceptron (MLP) on fault, and non-fault locations picked manually over seismic data to generate a single attribute referred to as fault meta-attribute probability. The generated fault meta-attribute probability has successfully illuminated geological discontinuities with efficiency. It is observed that in the western part of the study area, the fault probability has brought out thinned and sharpened fault images from seismic data, whereas the efficiency of this meta-attribute decreases towards the east due to the presence of noise around the fault’s positions. Based on these results, the interpretative approach followed in this work demonstrates the effectiveness of using machine learning and automatic methods on imaging geological structures from seismic data. We expect that these results can be used directly as input data in seismic hazard studies in this region.