Alireza Arabameri; Kalil Rezaei; Mohammadhossein Ramshet; Kourosh Shirani
Abstract
Landslide susceptibility and its risk assessment is the main part of landslide risk mapping. In this study, landslide susceptibility of Oliya's Padena in Semirom is mapped using artificial neural network. A total of 23 factors in relation to landslide in the region were initially characterized. The spatial ...
Read More
Landslide susceptibility and its risk assessment is the main part of landslide risk mapping. In this study, landslide susceptibility of Oliya's Padena in Semirom is mapped using artificial neural network. A total of 23 factors in relation to landslide in the region were initially characterized. The spatial location of landslide events was then determined by field study as well as aerial photo analysis. AHP analysis tends to 14 out of 23 parameters as the important factors for further steps. A total of 72 (70%) and 31 (30%) out of 103 detected landslide events in the study area were selected as training and validation data for neural network analysis, respectively. A multilayer perceptron back propagation algorithm with sigmoid as activation function was developed. The best topology was determined by using conventional criteria including mean square error, root mean square error, maximum absolute error and correlation coefficient. Results show that a 14-4-1 array is the optimum topology for landslide susceptibility zoning in the region. The weight of each input layer was estimated by frequency ratio. In order to map landslide, ROC graph and area under curve indices were used and the accuracy of output map was computed. Results from validation shows that area under curve for the obtained model is about 0.938 (93.8%) that is considered as high resolution prediction group. According to this study, a total of 29.61 square kilometers (93.25%) of the landslide areas is categorized in very high and high susceptible groups.
