Identifying Factors Affecting Landslides on Astara Road to Namin Tunnel Using the MLP Model

Document Type : Research Article

Authors

• Fariba Esfandiari Darabad ¹
• Mehrdad Vahabzadeh ²
• Zohre Shikhler ³
• Behrouz Nezafat takle ⁴

¹ Professor in Geomorphology, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran

² Masters graduate in Geomorphology and Environmental Management, Faculty of Social Sciences University of Mohaghegh Ardabili, Ardabil, Iran

³ Master's student in Remote Sensing, Faculty of Earth Sciences, Shahid Beheshti University, Tehran, Iran

⁴ PhD student in Geomorphology, Faculty of Social Sciences, University of Mohaghegh Ardabili, Ardabil, Iran

10.22067/geoeh.2024.87409.1475

Abstract

Landslides are one of the phenomena that cause significant damage, especially in mountainous areas. Therefore, it is essential to evaluate and identify the factors affecting the occurrence of landslides in these areas. The purpose of this research is to identify the factors affecting landslides along the Astara road to Nemin tunnel using an MLP model. The MLP model is one of the efficient neural network models that can solve complex problems. To identify the important factors in the occurrence of landslides, based on field studies, 8 factors have been identified, including: geology, vegetation, distance from the road, land use, slope, aspect, elevation, and recorded landslide locations. After pre-processing, all layers were entered into SPSS MODELER software, and the modeling was designed with 8 input neurons, 6 hidden layer neurons, and 1 output neuron. The results of this research showed that the weighted output of the MLP model assigned the highest weight to the geological layer (0.26), followed by the land use layer and distance from the road (0.14 and 0.13, respectively). For model validation, the AUC value was 0.948 for the training data and 0.962 for the testing data, indicating the model's high reliability in both phases. Therefore, it can be concluded that the geological factor has the greatest impact on the occurrence of landslides in the region compared to other factors. Finally, machine learning models and artificial intelligence are recommended for future studies on landslides and mass movements.
Extended Abstract
Introduction
Geomorphic phenomena are natural occurrences that manifest with specific spatial and temporal characteristics. According to the Geological Society, a landslide is defined as the downward movement of mass materials on sloping surfaces. Landslides cause significant damage worldwide each year and are recognized as one of the most critical geomorphological hazards. Natural factors such as earthquakes, rainfall, and snowmelt, along with human activities such as road construction, infrastructure development, and mining, exacerbate landslide occurrences, particularly in mountainous regions. One key strategy to mitigate damages caused by mass movements, particularly landslides, is the accurate and timely identification of areas with instability potential. Determining the factors influencing landslides and categorizing associated risks are crucial for developing effective mitigation strategies. The Astara–Namin tunnel road, a major route connecting Gilan and Ardabil provinces, is highly susceptible to landslides due to its unique geomorphic and geological conditions, its location in a mountainous region, and its specific climatic characteristics. Given the importance of this road and the associated slope hazards, especially landslides, conducting geomorphological studies to identify influencing factors is imperative. These studies can lead to effective planning and measures to reduce risks, prevent human and financial losses, and protect the local environment. This research aims to identify the factors affecting landslides along the Astara–Namin tunnel road using the Multilayer Perceptron (MLP) model.
Material and Methods
This study utilized various geospatial datasets, including a 1:100,000-scale geological map of Astara County for geological formations, a 1:50,000-scale topographic map for road layers, and a 12.5-meter resolution DEM from ALOS-PALSAR. Sentinel-2 satellite imagery was employed for land use and vegetation cover analysis, while GPS data was used to map landslide-prone areas. The Multilayer Perceptron (MLP) model is a type of artificial neural network inspired by the structure and functionality of the human brain. The MLP model is highly effective in solving complex problems by recognizing patterns and relationships in data.
Results and Discussion
The results indicate that the geological layer holds the highest weight (0.26), followed by land use (0.14) and distance from the road (0.13), suggesting that landslide occurrences along the Astara–Namin tunnel road are primarily influenced by these factors. The study classifies landslide-prone areas into four risk levels—low, medium, high, and very high—based on the Jenks natural breaks classification method. The area distribution for each risk class in the MLP model is as follows: very high-risk zones (23.14 km²), high-risk zones (47.2 km²), medium-risk zones (21.19 km²), and low-risk zones (0.9 km²). These findings demonstrate the model's capability in identifying critical landslide-prone areas along the studied route.
Conclusion
The presence of andesite and sedimentary rock layers, such as limestone and other fragile formations, contributes to slope instability under external dynamic forces. Excessive water intake from rainfall and snowmelt saturates the surface layers of slopes, transforming the soil into a semi-fluid state. Since the lower layers are more resistant, the upper layers tend to move downslope, leading to landslides. Additionally, human activities significantly influence landslide occurrences in the region. Extensive land use changes, particularly those that disturb natural vegetation and soil stability, exacerbate landslide risks. The MLP algorithm evaluation demonstrated high reliability in both training and testing phases, making it a valuable tool for identifying landslide-influencing factors. The distinguishing feature of this algorithm is its design, which mimics the human brain's neural networks, enabling it to solve complex problems effectively. This study highlights the necessity of incorporating advanced computational models such as MLP in geomorphological hazard assessments to enhance predictive accuracy and risk mitigation strategies.
 

Keywords

• Landslide Susceptibility Assessment
• MLP Neural Network
• Astara-Namin Road
• Geospatial Factors
• Risk Zonation
• Geomorphic Hazards

Main Subjects

• Sliding and mass movements

---

References

Abija, F. A., Nwosu, J. I., Ifedotun, A. I., & Osadebe, C. C. (2019). Landslide susceptibility assessment of Calabar, Nigeria using Geotechnical. Remote Sensing and Multi-Criteria Decision Analysis: Im-plications for Urban Planning and Development, 774-788. http://dx.doi.org/10.25177/JESES.4.6.RA.617

Alqadhi, S., Mallick, J., Hang, H. T., Al Asmari, A. F. S., & Kumari, R. (2024). Evaluating the influence of road construction on landslide susceptibility in Saudi Arabia’s mountainous terrain: a Bayesian-optimised deep learning approach with attention mechanism and sensitivity analysis. Environmental Science and Pollution Research, 31(2), 3169-3194.‏ https://doi.org/10.1007/s11356-023-31352-4

Asghari Saraskanroud, P., & Pirizi, A. (2023). Comparative evaluation of WLC, OWA, VIKOR and MABAC multi-criteria decision making algorithms in landslide risk zoning, case study: Givi Chai basin of Ardabil province. Natural Geography Research, 54(1), 65-94. [In Persian] https://doi.org/10.22059/jphgr.2022.333658.1007656

Cubito, A., Ferrara, V., & Pappalardo, G. (2005). Landslide hazard in the Nebrodi mountains (Northeastern Sicily). Geomorphology, 66(1-4), 359-372. https://doi.org/10.1016/j.geomorph.2004.09.020

Demuth, H. B., & Beale, M. H. (1992). Neural Network Toolbox User's Guide. Mathworks, Incorporated.

Entezari, M., & Kordavani, M. (2022). Landslide hazard zoning using GIS-based methods and radar data (Case study: Fereydoon Shahr). Journal of Natural Environmental Hazards, 11(33), 177-196. [In Persian] https://doi.org/10.22111/jneh.2022.38660.1810

Esfandiary Darabad, F., Rahimi, M., Navidfar, A., & Arsalan, M. (2020). Assessment of landslide sensitivity by neural network method and Vector machine algorithm (Case study: Heyran Road-Ardebil province). Quantitative Geomorphological Research, 9(3), 18-33. [In Persian] https://www.geomorphologyjournal.ir/article_122210.html

Ghateh, S., Malekzadeh, T., & Pourghasemi, E. (2024). Landslide risk zoning using satellite images (case study: Balharud watershed, Garmi city). Geography and Human Relationships, 6(3), 145-164. [In Persian] https://doi.org/10.22034/gahr.2023.394638.1860

Hong, C., Burney, J. A., Pongratz, J., Nabel, J. E., Mueller, N. D., Jackson, R. B., & Davis, S. J. (2021). Global and regional drivers of land-use emissions in 1961–2017. Nature, 589(7843), 554-561. https://doi.org/10.1038/s41586-020-03138-y

Jahandar, S., Aghagolzadeh, A., & Kazemi Tabar, S. J. (2018). Blind recognition of block codes in the presence of high noise using statistical methods. Modern Defense Science and Technology (Non-active defense science and technology), 10(4), 373-381. [In Persian]

Jouybari Moghaddam, Y., & Rostami, S. Q. (2018). Fusion of Markov Chain and SAX Method for Drought Probability Analysis (Case Study: Eastern District of Isfahan, Iran). Environmental Management Hazards, 5(3), 295-311. [In Persian] https://doi.org/10.22059/jhsci.2018.267316.414

Liu, X., Shao, S., & Shao, S. (2024). Landslide susceptibility zonation using the analytical hierarchy process (AHP) in the Great Xi’an Region, China. Scientific Reports, 14(1), 2941. https://doi.org/10.1038/s41598-024-53630-y

Mir, R. A., Habib, Z., Kumar, A., & Bhat, N. A. (2024). Landslide susceptibility mapping and risk assessment using total estimated susceptibility values along NH44 in Jammu and Kashmir, Western Himalaya. Natural Hazards, 120(5), 4257-4296. https://doi.org/10.1007/s11069-023-06363-6

Mirzai, F., Momeni, A. A., & Abdi, Y. (2024). Landslide hazard zonation along Khorramabad-Zal Highway in Lorestan province using analytical hierarchy process (AHP). New Findings in Applied Geology, 18(36), 264-282. [In Persian] https://doi.org/10.22084/nfag.2024.28365.1582

Mohammadi, N., & Sasanpour, F. (2021). Landslide and debris flow risk analysis in Haraz and Lavasanat roads. Water and Soil Management and Modelling, 1(4), 14-29. [In Persian]  https://doi.org/10.22098/mmws.2021.9138.1023

Mokhtari, M. (1993). Shirvan Astakhri landslide. Geographical Research Quarterly, 29, 117- 124. [In Persian] 

Nasiri, S. (2008). A perspective on landslides in Iran, a case study of slope instability on Haraz National Base Road. [In Persian] 

Raja, N. B., Çiçek, I., Türkoğlu, N., Aydin, O., & Kawasaki, A. (2017). Landslide susceptibility mapping of the Sera River Basin using logistic regression model. Natural Hazards, 85, 1323-1346. https://doi.org/10.1007/s11069-016-2591-7

Ranjithan, S., Eheart, J. W., & Garrett, J. H. (1995). Application of neural network in groundwater remediation under conditions of uncertainty. New Uncertainty Concepts in Hydrology and Water Resources, 133-140.

Rezaei Moghaddam, M. H., Mokhtari, D., & Samandar, N. (2021). Mass movements Modeling and Motion Sensitive Zone Management Using Statistical Algorithms in Ojan Chay Basin. Geography and Development, 19(63), 147-174. [In Persian]   https://doi.org/10.22111/j10.22111.2021.6190

Schalkoff, R. J. (1997). Artificial Nneural Networks. McGraw-Hill Higher Education. New York.

Zhao, H., Tian, W. P., Li, J. C., & Ma, B. C. (2018). Hazard zoning of trunk highway slope disasters: a case study in northern Shaanxi, China. Bulletin of Engineering Geology and the Environment, 77(4), 1355-1364. https://doi.org/10.1007/s10064-017-1178-1