Document Type : Research
Authors
1
Assistant Professor, Soil Conservation and Watershed Management Research Department, Khorasan Razavi Agricultural and Natural Resources Research and Education Center, AREEO, Mashhad, Iran.
2
1- Assistant Professor, Soil Conservation and Watershed Management Research Department, Lorestan Agricultural and Natural Resources Research and Education Center, AREEO, Khorramabad, Iran.
10.22092/wmrj.2023.361650.1531
Abstract
Introduction and Objective:
Landslide susceptibility zoning using different methods is one of the solutions for landslide management.Due to the wide range of occurrences of landslides, there is no single method to identify and prepare a zoning map for risk assessment. By applying scientific methods, a set of accurate tools is provided for the preparation and optimal use of the landslide zoning map, as well as the use of landslide prediction models, which reduces the problem of landslide risk identification and zoning. The aim of the upcoming study is to model the sensitivity of landslide occurrence using three methods of machine learning algorithm, random forest (RF), maximum entropy (ME) and support vector machine (SVM) algorithm. Then, the efficiency of these models is compared in zoning the sensitivity of landslides in Bar Neyshabur watershed, Razavi Khorasan province.
Materials and Methods:
For this purpose, the distribution map layer of landslides in the region including 73 landslides was prepared and randomly divided into two groups for model training (70%) and model validation (30%). Also, 16 factors affecting the occurrence of landslides in the studied area were identified according to the review of extensive sources and digital layers were prepared in the geographic information system. Then, the landslide hazard map was prepared based on the three mentioned methods. Next, in order to evaluate the accuracy of modeling and compare the efficiency of the models, the total quality index (Qs) was used.
Results and Discussion:
The results showed that the random forest algorithm method (RF) with Qs = 0.018 was chosen as the best model for the basin. Support vector models (SVM) with Qs = 0.014 and maximum entropy (ME) model with Qs = 0.013 are in the next priority, respectively.
Conclusion and Suggestions:
The random forest algorithm provides not only better results, but also more practical results; So that by matching the results obtained with the actual conditions through field visits, there is a very high match between the results of the landslide susceptibility zoning map using the random forest model and the actual evidence in the study area. So, assuming the concentration of management operations in high-sensitivity classes and choosing the random forest model as the best model, 75.5% of the area of the region will be removed from the management process and will result in the allocation of financial resources and less time.
Keywords
Main Subjects
)