@article {
author = {Khosravi, Mohsen and Jamali, Ali Akbar},
title = {Forecasting the Trend of Landslide Changes in the Northern Region of Quchan with Regard to the Factors Affecting Landslide Using Neural Network, Cellular Automata-Markov, and Regression Logistics},
journal = {Journal of Geography and Environmental Hazards},
volume = {7},
number = {3},
pages = {1-17},
year = {2018},
publisher = {Ferdowsi University of Mashhad},
issn = {2322-1682},
eissn = {2383-3076},
doi = {10.22067/geo.v0i0.71520},
abstract = {Introduction The landslide is one of the types of mass movements. It consists of the fast or slow movement of stone, soil or the sum of both on the slope downward. Many studies have been carried out by geomorphologists to understand the factors affecting mass movements. However, due to the complexity of such movements and the impact of several factors on it, there is still no definitive and sufficient result in this regard. Generally, many factors can be considered effective in mass movements such as the genus of formation, type of surface materials, topographic conditions (e.g., gravity, slope steepness and aspect), aggregate formation, tectonic condition (proximity to fault), climatic conditions (e.g., rain, surface water, soil moisture), and type of land use (e.g., urban, rural, agriculture, road, and so on). The objective of this study is to forecast the trend of landslide changes with regard to the factors affecting the landslide by neural network, cellular automata-Markov, and regression logistics. Area of study The studied area is part of the Tabarkabad basin, one of the main sources of the Atrak river basin located in the north of the city of Quchan and between the east longitudes 58 30 to 59 and north latitudes 37 to 37 20. Geologically, Tabarakabad Basin is part of the Kope-Dagh basin. The basin's boundary is determined by the mountains of Allah Akbar and the anticline of Zobar. The average height of this basin is 1885m, and the average slope is about 23%. In addition, about 50% of the slope of the basin is between 20 to 40 degrees. Materials and methods Landslides in the area Landslide dispersion map in all landslide studies from the identification, monitoring, zoning of sensitivity and especially risk assessment, and landslide risk is an integral part and as a basic and important layer. Landslide susceptibility map using multilayer perceptron (MLP) neural network model The MLP method in the land-change model is used because this model creates a network of neurons based on the input and output variables. The number of input neurons is equal to the number of variables (10 neurons) and the output neurons is equal to maps for each two classes (landslide and no landslide) as change trend probability. After a given number of repetitions, it is possible to get the lowest root mean square (RMS) error the network uses half of the data for training and with half the other data; it tests the network and gives the least error and most accuracy. After obtaining the highest accuracy of the training and testing of the network, the potential conversion map is provided. These maps determine the likelihood of converting applications into one another. It also describes the degree of effectiveness of each variable in the model. In this case, land use and geological type have the highest effect and the direction of gradient has the lowest role in the accuracy of 80% prediction. Then, a landslide prediction probability trend change map will be produced. Landslide susceptibility map using the logistic regression model By executing the model and using the probability map, we can identify areas with more potential, also with the proposed logistic function; the model can be used to measure the impact of each independent variable in the model. The positive coefficients have a greater effect and negative coefficients without effect in the model. Considering these land use coefficients, slope, elevation, and ultimately geological type of the area have the highest effect. Prediction of landslide changes using the Cellular Automata, CA-Markov method CA-Markov provides the images of classified landslides which are analyzed and output in the form of a probabilistic matrix of variations and an output image of the probability matrix of variations for the horizon. The probability change matrix indicates that the probability of each class of landed slip usage in the future will change to another Discussion and Results In the present study, landslide changes in a part of the Tabarkabad basin in northern Quchan in three periods of 2006, 2010, and 2016 with Google Earth satellite imagery, landslide-mapping using the multi-layer perceptron neural network, logistic regression, also the application of the Markov forecasting model and the modeling approach of land change modeler (LCM) of landslide changes were predicted for 2032. Comparison of landslide maps in the mentioned periods indicating an increase in the level of landslide areas. Considering topographic and geological characteristics and climatic conditions governing the area. In addition, the issue of intensifying land use change over the past decade has expanded in scope fluctuations. The results of geological maps of most of the area are located on the Sanganeh and Sarcheshmeh formations. The effect of land use on landslides has a direct effect. Factors such as the spread of rainfed land, the degradation of grasslands, and the development of drainage systems from rivers to slopes have been one of the most important reasons for landslide due to land use. The results showed that, due to the loosening of formations, land use change to poor pastures and agricultural lands, rainfalls increased. Also with Increasing, the altitude (above 1750 m) has reduced the risk of occurrence due to a decrease in gradient. Conclusion Finally, in order to evaluate and compare the results, a hybrid model of laminated perception neural network, logistic regression, CA-Markov model for modeling and predicting landslide changes. The predicted results of the three models indicate that the combined model of multi-layer perceptron neural network with a Kappa coefficient of 0.96 was better than the logical regression models and CA-Markov with Kappa coefficients of 0.86 and 0.72. Using a hybrid model of multilayer perceptron neural network for 2016, a prediction map was prepared and according to the acceptable accuracy of the model for the year 2032, a landslide prediction map was extracted.},
keywords = {GIS,Landslide,Neural Network,Markov,Forecast,LCM},
title_fa = {پیشبینی روند تغییرات زمینلغزش منطقه شمال قوچان با توجه به عوامل مؤثر بر لغزش به روش شبکه عصبی، اتوماتای سلولی- مارکوف و رگرسیون لجستیک},
abstract_fa = {زمینلغزش از عمده مخاطرات طبیعی است که هرساله باعث خسارتهای جانی و مالی فراوانی میشود؛ بدین دلیل پرداختن به موضوع شناسایی علل و دلایل ایجاد آن و ارائه راهکار جهت کاهش این خسارتها از اهمیت خاص برخوردار است. منطقه شمال قوچان در ایران به دلیل موقعیت جغرافیایی، ویژگیهای طبیعی و زمینشناختی و همچنین تغییر کاربری شدید زمین یکی از مناطق مستعد بروز زمینلغزش است. بهمنظور شناسایی و تعیین مناطق حساس به لغزش با کمک گرفتن از عوامل مؤثر در آن با استفاده از روش شبکه عصبی، روش آماری مارکوف و رگرسیون لجستیک این پژوهش انجام شد. نقشه زمینلغزش موجود با استفاده از تصاویر ماهوارهای گوگل ارث برای سالهای 2004،2010 و 2016 تهیه شد و به دو کلاس، محلهای لغزش رخداده و رخ نداده طبقهبندی شد. مقایسه نقشه شبیهسازیشده سال 2016 هر یک از مدلها با نقشه طبقهبندیشده واقعی نشان داد که شاخص کاپا در نقشههای حساسیت لغزش ایجادشده توسط مدل شبکه عصبی چند لایه پرسپترون، مدل مارکوف و مدل رگرسیون لجستیک به ترتیب برابر 0.96، 0.72 و 0.86 است. این بدان معنی است که مدل شبکه عصبی پرسپترون مدل مناسبی برای پیشبینی تغییرات زمینلغزش در این منطقه است. نتایج نشان داد، با سست شدن سازندها، تغییر کاربری اراضی به سمت مراتع فقیر و اراضی کشاورزی دیم، افزایش شیب خطر رخداد زمینلغزش زیاد شده است.},
keywords_fa = {GIS,زمین لغزش,شبکه عصبی,مارکوف,پیش بینی,LCM},
url = {https://geoeh.um.ac.ir/article_32522.html},
eprint = {https://geoeh.um.ac.ir/article_32522_12db723a283e56acf5a424fc44095dd3.pdf}
}