Landslide susceptibility mapping of the Mediterranean coastal zone of Morocco between Oued Laou and El Jebha using artificial neural networks (ANN)

The goal of this study was to experiment artificial neural network (ANN) classifier on various available physical factors in the study area to produce a reliable landslide susceptibility map. The mapping of landsides is classically established through the identification and analysis of hillslope instability factors. Even if a variety of approaches use these analyses with geographic information system (GIS) performances to carry out a good result, there is no satisfaction because of the complexity of the landslides encountered in the field. In the present study, landslide susceptibility models were produced by using multilayer perceptron (MLP) ANN in the Mediterranean Rif coastal zone of Morocco. This was established in the following steps: (i) production of landslide inventory map; (ii) production of the hillslope factors, twenty factors composed of geology, geomorphometry, proximity, and thematic data derived from satellite imageries; (iii) extraction of vector model to be used to train ANN, construction of ANN models; (iv) validation and evaluation of results. The results of the prediction models were evaluated by the receiver operating characteristic (ROC) curves. The obtained area under the curve (AUC) values are greater than 0.90, indicating that the models are quite accurate. The visual comparisons between landslide susceptibility maps and the input factor maps show that roads and geology are the most important factors influencing five types of mass movements (complex, slide, flow, and rockfall). The success of this work will be helpful to expand this method to the whole Rif mountains in Morocco.