ارزیابی و پهنه‌بندی خطر زمین‌لغزش با استفاده از فرآیند تحلیل شبکه و شبکه عصبی مصنوعی (مطالعه موردی: حوضه آذرشهر چای)

Introduction
Mass movement is as relocation large amounts of masses of soil, stone or a combination of them down the hillside is, in effect of force gravity. In the process and how to mass movements in hillside are presented multiple classifications. Carson and Corby, mass movements materials on hillside have divided to three floors slip, flow and creep (Moghimi et al., 2008). Since predict the time and location of the landslides are out of the current human knowledge. Therefore, for expressing sensitivity slope to landslide hazard zonation is in different areas(Shadfar et al., 2007). Of the most common of natural hazards at Sahand mountainous masses are  a landslide. However, despite such abundance, exhaustive research has been done in this case. In these mountains, are harsh climatic conditions and context's sovereignty of materials susceptible to landslide such as pyroclastic materials and old alluviums is considered of landslide prone areas of the country. Azarshahr Chay basin as one of the drainage basins of Sahand mountain due to steep hillside's soil and surface materials not consolidate the lack of full-scope protection by vegetation and being active different processes over the years, feet walls of valleys cut by the waters flowing and in recent decade's manipulation human, the environment one of the area's susceptible mass movements(Bayati khatibi, 2011).
Methodology
In this study, landslide hazard zonation under study was checked at eight factors influencing to include: slope, aspect, lithology, distance from the road, distance from the fault, distance from the river, elevation and land use. These factor's maps prepared by ARC GIS software, and for zoning has been used in artificial neural networks and analysis network process. For this purpose the topographic map in scale 1: 50000 , geological map 1: 100000, 30-meter SRTM DEM been used the study area. By using satellite images Landsat 8 ETM+ 2014 (34 row; 168 Path), Google Earth software and field studies, were diagnosed 35 slide point. Then, the coordinates of the slide point transferred to the software Arc GIS landslide distribution map was produced in the context of these Apps. Also, the study  points of non-slide, for use in artificial neural network training and testing are provided in the slope of less than 5 degrees.
Results and discussion
As mentioned, one of the models used in this study is network analysis. For preparation, the map zoning the landslides uses analysis network process model, you first need effective factors coefficients calculated in landslides in the study area. Then, use the coefficients obtained, proceed to Creating map landslide susceptibility was based on the analysis network. To do so, first in ARC GIS information layers that already been prepared and Digital to become raster formats or network and then reclassified and eventually coefficients obtained from ANP model in the Raster Calculate were imposed on the aforementioned layers and finally zoning map were obtained from a network analysis model. And map was classified in five classes: very high, high, medium, low, very little.
The neural network model, after determining the basic structure of the neural network and providing the information needed to train designed the neural network. Also, to achieve an acceptable error, is ready for the network, to do the logical analysis of information that has not been encountered previously and done forecasting and simulations the necessary. For this purpose, using the weight of the final stage of the training network, The total area, including 56040 pixels and each of pixel has eight features a landslide. Enter the network. After performing network analysis on this data, for each pixel was obtained value between zero and one. By category the values obtained from the network, Areas to different parts divided of opinion landslide risk. Eventually, hazard maps by a margin of 0.2 were classified to five zones of very low, low, medium, high and very high.
Conclusion
In this study, we examined the 8 factors the landslide hazard zonation. These factors include slope, aspect, lithology, land use, elevation, distance from the road, distance from the fault, and distance from the river, which results from these studies are as follows. The results of the network analysis process as one of the multi-criteria decision-making models, the most important factors in the occurrence of landslides Azarshahr Chay basin in order of priority were identified as follows: 1. Lithology, 2. Land use, 3. Slope, 4. Elevation, 5. distance from fault, 6. distance from river, 7. Aspect, 8. distance from road.
For landslide hazard zonation using Neural Network, In the first stage of training in order to avoid increasing the error, Each of artificial neural network parameters (number of repetitions, then learning and the number of neurons in the hidden layer) is determined based on trial and error. Then 1-15-8 structure of meaning 8 input neurons, 15 neurons in the hidden layer and one output neuron have been prepared landslide maps; the results show that is located 5.49, 32.61, 32.05, 23.22, 5.73 percent of area are at risk classes very low, low, medium, high and very high. For landslide hazard zonation using a network analysis process, the formed paired comparison matrix factors and then was calculated weight each factor. Eventually, using the functions overlaps in geographic information system software for landslides in was prepared the final maps Azarshahr Chay the basin. The results show that is located 7.13, 28.44, 37.13, 23.14, 3.27 percent of the area are at risk classes very low, low, medium, high and very high. The results of the analysis network process and artificial neural network, using the coefficient of kappa statistics shows that artificial neural networks with coefficient 0.74 compared to network analysis process with coefficient 0.72 more accurate is in predicting the risk of landslides in Azarshahr Chay basin.