Zonation of unstable slopes with respect to the debris flows using random forest algorithm (case study: Basin Tngrah Golestan Province)

Abstract
In 2001 and 2002, during a sudden rainfall, the area in question experienced floods with irreparable flow of debris. . the tangrah Catchment in Golestan Province is one of Dough River sub- basins. occurrence and Storm rain falls in 2001-2002, triggered a severe debris flow which left many casualties and economic losses. One of the important causes of sediment flow rock deposits is slippery masses that occur in the basin. Therefore, identifying areas that are more vulnerable to debris movements can be effective in planning to deal with reducing the effects of these events. In the present study, for a more detailed analysis, a landslide hazard map has been prepared using a random forest algorithm in this area. Factors affecting the occurrence of landslides in the strait area in relation to the debris flow, including parameters such as slope, slope direction, altitude, curvature amplitude, lithology, total annual rainfall, land use, distance from waterway, distance from fault and distance from Are ways of communication. Using Mat LAB R2020a software, 70% of this data was randomly selected for training and the rest was used for validation. Evaluation of the results obtained from the model of random forest algorithm has been created using a coefficient of determination of 0.88 and an error of 0.27. Prioritization of input factors by the algorithm indicates the greater importance of the factors of slope, height and distance from the road in the final forecast. The secondary validation is by the ROC criterion and the area below the curve is 0.93, which indicates the estimated accuracy of the created model. Based on the zoning, the results show that 16, 15, 11, 17 and 40% of the area are in very low, low, medium, high and very high classes, respectively.
Keywords: Zonation, landslide, debris flow, Tangarah basin, Golestan province
Introduction
Debris flow is one of the most common geological hazards, often starting with landslides, and the potential energy of a slippery mass can be rapidly converted into kinetic energy. Deposits can lead to catastrophes that pose a serious threat to the lives and property of individuals and to economic development. Landslide is an integrated and often rapid volume of sedimentary material along slopes (Mahmoudi, 2007: 38). According to the definition of the Engineering Geological Society, landslide is the movement of a mass of material on a downward slope (Nasiri, 1383: 3). Therefore, these phenomena are one of the most important natural hazards that have serious casualties and financial losses around the world, especially in mountainous basins (Vilanova et al., 2010: 383) Landslides are of two types: Landslides in about one Meters is called shallow and a few meters deep. Shallow landslides contain a lot of water that occurs after heavy rainfall, but deep landslides require a longer time. When the earth's water level rises enough, the earth's block becomes unstable, causing landslides that often occur after the heaviest rains. Hence it will be a mechanism for creating a deposit (Takahashi, 124: 2007). The debris flow is actually the motion of old and new slips, and the potential energy of the slippery mass can be quickly converted into kinetic energy. Material flows can lead to catastrophes that pose a serious threat to the lives and property of individuals and to economic development. Many countries suffer from the serious risks of a deposit flow (Liang et al., 2012: 95).
Methods and material
Random forest is a modern type of rootstock that includes a host of classification and regression trees. The RF predictor model is based on averaging the results of all relevant decision trees and performs high-accuracy classification for many sets (Ebrahim Khani et al., 2011). The most important feature of random forest is its high performance in measuring the importance of a variable to determine what role each variable plays in predicting the response. The research has been prepared to prepare a map of sensitivity to debris flows and landslides in which 10 effective factors in amplitude instability have been used. After preparing 10 independent factors and landslide data, the coding of random forest algorithm in Mat LAB R2020a environment was used.

Results and discussion
To determine the appropriate number of trees, using the mean square error (MSE) criterion, first some initial values for the number of trees were determined and then the model was implemented. By examining the mean error rate and the coefficient of determination for training, the optimal model with the lowest error rate was designed. The obtained model with a coefficient of determination of 0.88 and the square of the mean squared error was 0.27 for the training stage. In order to achieve a logical and appropriate model for the strait catchment, the ROC curve has been used. ROC curvature is one of the most complete methods in providing feature determination, probabilistic identification and prediction of systems, which slightly reduces the accuracy of the model. In this rock curve evaluation, the higher the rock curve surface, the higher the model accuracy.
Studies have shown that factors such as slope, height and distance from the road have a very effective role in creating landslides in the region and factors such as distance from the waterway and distance from the fault and the direction of the slope have the least impact on landslides, respectively. slope factor is the first factor affecting instability in the region. Most debris flows originate from discrete or distributed source areas where slopes steeper than 30 ° are covered by debris or soils with low adhesionConclusion
The results of the present study are consistent with the results of researchers such as Pour Ghasemi et al. (2015) in Mazandaran and Mohammadi et al. (2017), in a part of Golestan and Talebi et al. (2016) in Sardarabad watershed, Golestan province .