Evaluation of MLP and RBF Methods for Hazard Zonation of Landslides Triggered by the Twin Ahar-Varzeghan Earthquakes

Seismic landslides are among the phenomena posing substantial financial and life losses to human societies. Although many studies have been conducted on detection and description of landslides in general, seismic studies of landslides, particularly in Iran, are in their early stages. In this research, two types of artificial neural networks including multilayer perceptron (MLP) and Gaussian radial basis function (RBF) were used for the zonation of seismic landslides at a 1:50,000 scale. The earthquake-stricken area of the twin Ahar-Varzeghan Earthquakes was selected as the modeling case. First, data including disturbance distance (distance from river and road), ground strength class, moisture content, shake intensity, slope angle, and vegetation were collected using the aerial photo, LANDSAT images, topography and geology maps, and site investigations. Next, these data were digitalized using the geographic information system and weighted based on the density of landslides occurred in the study area. Finally, by applying two artificial neural networks (MLP and RBF), the study area was zoned and the accuracy of these methods was evaluated using the occurred landslides triggered by twin Ahar-Varzeghan Earthquakes. Considering that the present research is conducted using a larger number of parameters and more accurate analyses, the obtained results showed that the sum of density ratio values of the occurred seismic landslides in the study area for “medium hazard”, “high hazard”, and “very high hazard” zones using the zonation maps prepared by MLP and Gaussian RBF is 4.054 and 2.032, respectively; with MLP providing the optimum response. Finally, comparing these two methods based on hazard zonation maps and quality sum criterion, it can be concluded that MLP technique involves a higher precision compared to the Gaussian RBF technique in the study area. © 2017, Springer International Publishing Switzerland.