نوع مقاله : پژوهشی

نویسندگان

• اسدالله حجازی ¹
• محمد حسین رضایی مقدم ²
• عدنان ناصری ³

¹ گروه ژئومورفولوژی دانشکده برنامه ریزی و علوم محیطی دانشگاه تبریز

² استاد/دانشگاه تبریز

³ گروه جغرافیای دانشگاه پیام نور مرکز سنندج

چکیده

هدف تحقیق حاضر بررسی و مقایسه دو مدل شبکه عصبی مصنوعی و مدل تاپسیس در پهنه‌بندی خطر زمین لغزش در منطقه پایین‌دست سد سنندج است. این کار با استفاده از نرم افزار ARCGIS، زبان برنامه نویسی پایتون و مدلهای شبکه عصبی مصنوعی و تاپسیس صورت گرفت. بدین منظور از 9 لایه ورودی در پهنه بندی خطر زمین لغزش استفاده شد. نقاط لغزشی و غیر لغزشی منطقه با استفاده از تصاویر ماهواره ای، مشخص گردید. از وزن یابی درونی در تعیین وزن لایه ها استفاده شد. در مدل شبکه عصبی داده ها با استفاده از یک شبکه پرسپترون چندلایه با الگوریتم یادگیری آدام آموزش دیدند. ساختار شبکه دارای 9 نرون در لایه ورودی، 30 نرون در لایه میانی و 1 نرون در لایه خروجی است. در مدل تاپسیس پس از بی مقیاس سازی ماتریس تصمیم از روش آنتروپی شانون برای وزن دهی به معیارها و به منظور تعیین فاصله نسبی از ایده آل مثبت و منفی از فاصله اقلیدسی استفاده شد. پس از آماده سازی مدلها، منطقه مورد مطالعه با 970 کیلومتر مربع با 9 متغیر ورودی که تبدیل به داده های رستری به پیکسل های 30*30 شدند تحلیل شد. نتایج تحلیل به صورت نقشه ای با پنج طبقه خطر زمین لغزش برای هر مدل ترسیم گردید. بعد از اعمال 5 روش محاسبه میزان خطا جهت صحت سنجی مدلها مشخص گردید مدل شبکه عصبی پرسپترون دارای خطای کمتر و انطباق بیشتری است و با جغرافیای منطقه سازگاری بهتری دارد.

کلیدواژه‌ها

• "پهنه بندی خطر"
• "زمین لغزش"
• "شبکه عصبی"
• "تاپسیس"
• "حوضه آبخیز قشلاق سنندج"

20.1001.1.23833254.1399.7.24.4.2

عنوان مقاله [English]

Landslide hazard zoning using artificial neural network models and TOPSIS downstream of Sanandaj Dam

نویسندگان [English]

• asadollah hejazi ¹
• mohammadhossein rezaeimoghaddam ²
• adnan naseri ³

¹ Department of Geomorphology, Faculty of Planning and Environmental Sciences, Tabriz University

² Professor, Department of Geomorphology, Faculty of Planning and Environmental Sciences, Tabriz University.

³ Geography Department of Payame Noor University, Sanandaj

چکیده [English]

1-Introduction
The purpose of this study is to select the best model and identify landslide risk areas in the downstream basins of Sanandaj Dam. Every year, mass movements in the region cause damage to roads, natural resources, farms and residential areas, and increase soil erosion. Kurdistan province, with its mostly mountainous topography, high tectonic activity, diverse geological and climatic conditions, has the most natural conditions for mass movements. According to the available statistics, this province is the third province in terms of landslides after Mazandaran and Golestan. (Naeri & Karami, 2018). The Gheshlagh River Basin is a mountainous region with a north-south trend. In terms of construction land, it is located on the structural zone of Sanandaj-Sirjan. The study area with an area of 970.7 square kilometers is located downstream of Sanandaj Dam. The city of Sanandaj is located within the basin. Effective parameters for landslides according to the type of climate and morphological processes are provide in the geography of the region.
2-Methodology
The present study includes five stages of research background and data collection, preparation of information layers, implementation of artificial neural network and TOPSIS models, preparation of landslide Hazard zoning map in gheshlagh basin with the mentioned models and validation test of the models. In this study, nine effective factors for landslides, including slope, slope direction, fault distance, road distance, waterway distance, lithology, land use and precipitation were used .Using Google Landsat 8 ETM satellite imagery, Google Earth software identified 237 slip points. Then, the coordinates of the slip points transferred to the Arc GIS software and a map of the landslide distribution area in this environment was prepared. In addition, in this

 

study, 89 non-slip points were prepared for use in the training and testing stages of Persephone neural network inside slopes less than 5 degrees. Artificial neural networks are made of a large number of interconnected processing elements called neurons that act to solve a coordinated problem and transmit information through synapses. Neural networks begin to learn using the pattern of data entered into them. Learning models, which is actually determining their internal parameters, based on the law of error correction. In this method, by correcting the error regularly, the best weights that create the most correct output for the network identified. The neurons are in the form of an input layer, an output layer, and an intermediate layer. TOPSIS is a very technical and powerful decision-making model for prioritizing options by simulating the ideal answer. In this method, the selected option should be the shortest distance from the ideal answer and the farthest distance from the most inefficient answer,) Dong, 2016). In the artificial neural network model, the middle layer selected by default. Percentage70 of the landslides occurred for neural network training and the remaining 30% as reference data used to test and calibrate the model. Data trained using a multilayer perceptron network with Adam learning algorithm. The final structure of the network has nine neurons in the input layer, 30 neurons in the middle layer and 1 neuron in the output layer. In the TOPSIS model, after scaling the decision matrix, Shannon entropy method used to weight the criteria and to determine the relative distance between the positive and negative ideals of the Euclidean distance.
3-Results and Discussion
The final structure of the network has nine neurons in the input layer, 30 neurons in the middle layer and 1 neuron in the output layer. In the TOPSIS model, after scaling the decision matrix, Shannon entropy method used to weight the criteria and to determine the relative distance between the positive and negative ideals of the Euclidean distance. After creating the raster layers of each index in the TOPSIS model, a vector-point layer created that has one row per pixel and one column per index, thus creating a matrix with dimensions of 9 by 1078555. The operation of Salavatabad fault in the east of the basin has caused Horst and Graben in the region. The significant difference between the height of the mountain unit and the riverbed has caused hazards and the transformation of landforms in the region. In both models, the western part of the basin is in a very high-risk zone, and housing and mass movements threaten agricultural land in these areas. The western outskirts of Sanandaj, which is located in the center of the basin, also affected by numerous landslides and classified in the high and very high danger zone.



4- Conclusion
The downstream area of Sanandaj Dam is one of the most active areas of Kurdistan province and the west of the country in terms of human activities. Out of a total of 970 square kilometers, the area under study, according to the neural network model, is about 31 percent and the TOPSIS model is 30 percent of the area within the optimal areas for human activities. In addition, according to the neural network model, about 39% and the TOPSIS model 42% of the region are in the range of undesirable and very undesirable areas. The results show that the study area in general has a high potential for landslides. Dangerous areas are located mainly in the west and southwest of the constituency. These areas correspond to the mountain unit, rainfall of more than 385 mm and high slope. Rainfed agriculture and rangeland with medium-sized canopy are widespread in this area. These areas are also located on the k8, kp1 and PE geological units. Comparison of the results of risk zoning validation in the model shows that in this area, the perceptron neural network model has a better accuracy than the TOPSIS model.
Keywords: Hazard zoning, Landslide, Neural network, TOPSIS, Sanandaj Gheshlagh Watershed
5-References
Dong, S. (2016). Comparisons between Different Multi-Criteria Decision Analysis techniques for Disease Susceptibility Mapping. Student Thesis Series INES. Department of Physical Geography and Ecosystem Science Lund University Sölvegatan. Sweden 12 S-223 62.
Geological Map Description Sanandaj 1: 1000000. (1990). Geological Survey if Iran. Tehran. Iran
Naeri, R. Karami, M. (2018). Integration of Analytical Zoning Risk of Bijar Lanslade Occurrence, Journal of Engineering Geology, 12(1), 153-182.
 
 
 

کلیدواژه‌ها [English]

• "hazard zoning"
• "landslide"
• "neural network"
• "Topsis". "Sanandaj Gheshlagh watershed"