On the viability of Neural Networks for landslide susceptibility mapping in the Rif, North of Morocco

Landslides are among the deadliest natural disasters in the world. With the rising effects of climate change, extreme events responsible for triggering landslides are becoming more and more frequent. This paper uses an Artificial Neural Network (ANN) to develop the first deep learning geospatial model for landslides in the Rif (North of Morocco), and compares its performance in landslide susceptibility mapping with logistic regression. The models input parameters can be derived from free geospatial data, which makes this approach efficient and applicable over large areas. The models were validated using landslide data derived from geological maps of the Rif area. We noted an accuracy of 0.85 for the ANN model and 0.69 for the logistic regression model respectively. A susceptibility index map was generated in the Aknoul region in the Rif to illustrate the ANN results. We found that areas with very high susceptibility index are affected the most by landslides.