Modeling the hillside movement in the area of Sattarkhan dam reservoir using by predictive models Logistic Regression and Neural Network

Slope instabilities are considered as one of the major dangers to human activities which occurred in natural slopes and the slopes made by humans. This study aimed to identify the factors affecting the occurrence of slope instability using logistic regression models and artificial neural network in Ahar Sattar khan dam basin. The results of statistical models to determine the potential areas of instability and ultimately create a hazard zonation map for the study area. In this regard, the most important factors in landslides such as slope, aspect, elevation, rainfall, distance from the road, fault and drainage, land use and exploration and peculiarities of each of them were identified. Standardization base of histogram model is by cutting the classes of each layer with occurred landslides. The models show that the very high-risk area in the neural network and logistic regression are 724 and 5/56 per cent respectively which cover the areas close to Sattar khan dam, mainly including the lithology of these areas located in the regions with lower resistance. Besides, Statistical methods Logistic prove that faults and lithology  have an immense impact on the occurrence  of landslide in this area. The ROC index value for neural network and logistic regression models are 0/85 and 0/81. So it can be said neural network model for zoning of landslides is more efficient, so any planning and construction must be compatible with the conditions of geomorphology and geology of the area leading to as least human and financial losses as possible.
Extended Abstract
1-Introduction
Evaluation of slope instabilities such as many of the most complex issues of environmental hazards are considered as the important factors in the occurrence of slope instability, mostly due to the variety of issues. Despite the uncertainty caused by being vague, incomplete and ambiguous terms and concepts related to parameters such as geology, hydrology, tectonics, vegetation, rainfall, erosion, temperature fluctuations, the impact of ice and the instability domain make the need for accurate and convenient methods seem reasonable. This method is quite small and the effect of each independent variable on the dependent variable to quantify and anti-log coefficients and coefficients are specified. Considering the background of selected models, the area of Sattar Ahar dam, having a lot of  slope motion phenomena, is regarded as a prone area, therefore, applying this model, we can make sure about the validity of the findings. The aim of this study is to evaluate the effect of risk factors using statistical models of logistic and neural network in the catchment area around the dam of  Sattarkhan Ahar in East Azerbaijan.
2- Materials and Methods
study area around the Sattar khan dam with an area equal to 48/147 square kilometers, is located 11 km west of Ahar. The basin is a part of Ahar-chai River Basin. In the present study, artificial neural network and logistic regression models were used to map  the landslide hazard around the area of Ahar Sattar Khan. Neural network model is a computational mechanism which is able to capture data and calculate a set of new information. This model is superior to other methods, including artificial neural network that is independent from the statistical distribution of data and does not require special statistical variables. Compared to other statistical analysis, logistic regression model requires fewer assumptions and normal distribution of variables, non-linear relationship between dependent and independent variables. This method is quite small and the effect of each independent variable on the dependent variable to quantify and anti-log coefficients and coefficients are specified. To use logistic regression and neural network models, it is required to use a set of data includeing both data and track their documents. Data, documents, library studies, topographic maps with a scale 1/50000, geological map at a scale of Ahar 1/250000 registered land slide points on satellite images Quick Bird, Google Earth Systems and data from earth observation and photography of occurred landslides. Digital elevation layer of polyester is the one with a pixel size of 30 meters and meteorological data and climatological Water Authority. In this study, a total 10 factors, including mass movements, lithology, faults, elevation, slope, aspect, land use, distance ftom road, maps of precipitation and drainage network congestion and study have been used.
3- Results and Discussion
Effective layers of factors in the instability range layers should be made to map landslide hazard, the factors are suggested to be more than 10.  In this study, two methods of artificial neural network and logistic regression models were used to map the zoning and sensitive areas around the reservoir in the area of  Sattar Khan Ahar. The factors contributing to the occurrence of occurred landslides in the study area were prepared to enter the models. After extracting all the layers on the next phase, each extracted layer and the layer of mass movements are crossed. Mapping data for the final application and verification of each map was done by ROC showing high accuracy. In this study, the distance from river, lithology, faults and slope play an important role in landslides of the area. Besides, the data from 1500 pixels and non-slip sliding grid are used to train and test, in which 1000 pixels were used for training and 500 pixels network testing.
4- Conclusion
According to zoning map output,  the danger of probable slope motions in both models areas is in the range of  high risk.  Therefore, the potential for instability in the region is very high and it is urgent to plan and control the measures. High-risk areas in logistic model is 5/56 percent and 7/24 percent in the neural network model. As a result, it can be said that in addition to natural factors, human factors, including incorrect road, have an important role in the occurrence of unstable slopes. It is necessary to avoid land use change and ecosystems to reduce the risks and increase the stability of slopes. Technical review areas of potential mass movements must fit the model number of criteria. Since the model used in this study have a lot of criteria and sub-criteria, the findings have high accuracy.