نوع مقاله : مقاله پژوهشی
نویسندگان
1 بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی لرستان، سازمان تحقیقات، آموزش و ترویج کشاورزی، خرم آباد، ایران
2 بخش تحقیقات جنگل‌ها و مراتع، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی خوزستان، سازمان تحقیقات، آموزش و ترویج کشاورزی، اهواز، ایران
3 بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات و آموزش کشاورزی و منابع‌طبیعی خراسان‌رضوی، سازمان تحقیقات، آموزش و ترویج کشاورزی، مشهد، ایران
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Introduction
Landslide is one of the natural hazards that causes a lot of human and financial damage every year in mountainous, rainy and seismic areas. Mass movements play an effective role in destroying communication roads, pastures, mountainous areas and causing erosion and sedimentation in watersheds. Identifying landslide-prone areas through risk zoning with appropriate models is unique of the main methods in reducing potential damage and hazard managing. Landslide susceptibility map preparation is known as the cornerstone of landslide research and is used as a management tool in times of crisis. Considering that the identification of landslide sensitive areas based on traditional methods and expert opinions is not accurate enough, the use of modern machine learning methods such as the support vector machine method seems to be necessary and necessary. The purpose of this research is to spatially model landslide susceptibility using two methods: generalized linear model (GLM) and support vector machine (SVM) and compare the efficiency of these models in zoning landslide susceptibility in Karganeh Watershed, Lorestan Province.
Materials and Methods
Karganeh Watershed is one of the large sub-watersheds of Khorramabad with an area of 294.2 square kilometers. The minimum height of the this watershed is 1300 and the maximum is 2700 meters, and 60% of the area of this watershed has a slope of more than 12% (relatively high slope and more).The research method in this study is applied in terms of purpose and in terms of descriptive-analytical nature, library methods, field visits and modeling are used. On behalf of this goal, the distribution map layer of landslides in the watershed including 95 cases of landslides was prepared and separated into two sets for model training (70%) and model validation (30%) randomly.
Also, 16 causes disturbing the happening of landslides in this watershed were selected permitting to the review of sources and the usage of principal component analysis (PCA), Tolerance and VIF tests. Digital layers of effective factors in geographic information system were equipped. Slope factors, slope direction, elevation classes, geology, drainage network (distance from the river), road (distance from the road), fault (distance from the fault), topographic indicators (river power index (SPI), topographic moisture index (TWI) and Slope length index (LS)), geomorphological indices (topographic position index (TPI), topographic roughness index (TRI) and power vector index or surface roughness measurement (VRM) of land use, distance from the village, and rain lines as the most effective factors Landslide occurrences in Karganeh Watershed were known. Then, the landslide hazard map was prepared based on the two mentioned methods in the ModEco software environment. Next, in order to evaluate the accuracy of the modeling and compare the efficiency of the models, the relative performance recognition index (ROC) was used.
Results and Discussion
By correcting the landslide data provided by the General Directorate of Natural Resources with the help of Google Earth satellite images and field visits, 95 landslides were identified, which cover an area of 1483.24 hectares of this watershed. Established on the fallouts of the maximum likelihood diagram, geological, land use, slope, topographic roughness index (TRI), slope length and slope direction are the best significant factors inducing the event of landslides in Karganeh Watershed. The results showed that the support vector machine (SVM) method with ROC equal to 0.913 was chosen as the best model for the basin. The generalized linear model with ROC equal to 0.803 also showed high efficiency in terms of landslide susceptibility evaluation. Permitting to the results of the support vector machine, about 19.3% of the Karganeh Watershed is in the high and very high hazard class of landslide happening. Based on the landslide susceptibility map with the support vector machine model, the villages of Jamshidabad Haider, Milmilk, Garmabala Bala, Bene Soura, Mahmoudabad Bala, Skin Bala, Cheshme Papi, Dareh Qasim Ali and Sheikh Haider are at high and very high hazard of landslides. It were located About 117 kilometers of communication roads in this area were classified as high and very high hazard.
Conclusion
Established on the fallouts of this study, the maximum entropy algorithm provides practical results in order to zone the hazard of landslides in the Karganeh Watershed; By matching the obtained results by the existing real conditions, there is a very high agreement between the fallouts of the landslide hazard zoning map using this model and the real indication in this watershed. Assuming the concentration of management operations in high-sensitivity classes, about 70% of the area of the region will be removed from the management process and will cause the allocation of financial resources and less time. The implementation of landslide management programs based on the results of this research on a local and regional scale can explain the difficulties of domain instability and lead to the development of watershed management actions and the sustainability of the development of the Karganeh Watershed.
کلیدواژه‌ها [English]
)