Landslide susceptibility mapping by using multi-layer perceptron neural network model of back error propagation (Case Study: Bar Basin of Neyshabour)

With regard to the capabilities of the artificial neural networks, applying them in a variety of engineering and geologic fields has been developed remarkably. In this study, for landslide zoning across Bar Basin in Neyshabour, the multi-layer perceptron model of back propagation (BP) were used. In order to assess the neural network created, data of 32 landslides were entered into the system. This database include information about slopes, aspects, lithology, digital elevation model (DEM), map isohyets, distance from the fault, and landuse. To feed these data to the created neural network, they were normalized based on the highest rate of each data in the database between zero and one. Then, normalized data were fed to a three-layer feed forward perceptron neural network with back error propagation algorithm. The abovementioned data were primarily trained in the network and then were tested. The final structure of the network has seven neurons in the middle layer and one neuron in the external layer. Among them, 80 Percent of the data were used for training and the remaining 20 Percent for tests. Finally, considering the external weight, zoning map of landslide were drawn in five ranks from very high risk, high risk, medium risk, low and very low. The results show that the geologic structure developed due to the grey marns with lime layers (Delichai construction) and also the faults of the tectonic area caused the Bar basin to have a high capability in terms of landslide formation.