Document Type : Research Article
Authors
1
Department of Physical Geography, Faculty of Geographical Sciences and Planning, University of Isfahan, Isfahan, Iran
2
Department of Marine Geology, Faculty of Marine Natural Resources, Khorramshahr University of Marine Science and Technology, Khorramshahr, Iran
Abstract
Landslides are natural hazards that pose significant risks to human lives and the environment. Landslide susceptibility maps are vital tools for planning, management, and risk-reduction strategies. Various methods exist for generating these maps, but their effectiveness varies across studies. This research aimed to develop a landslide susceptibility map for the Shahid Abbaspour Dam watershed using an Artificial Neural Network (ANN). Landslide data points were randomly split into a 70% training and 30% testing dataset. Fifteen influencing factors were selected as model inputs, including elevation, slope, aspect, curvature, Stream Power Index (SPI), Topographic Wetness Index (TWI), distance to roads and rivers, geology, soil texture, land use, and precipitation. Model validation was performed using accuracy metrics and statistical indices (RMSE, Cohen’s Kappa, MAE, and R²). The ANN model demonstrated strong predictive performance, achieving an accuracy of 0.8543. The study identified rivers as the most critical factor influencing landslide susceptibility in the area, followed by slope and geological formations. The highest susceptibility zones were found in the southwest and in a small northern section of the watershed. The resulting map provides valuable support for land-use planning and sustainable development strategies aimed at mitigating landslide risks in the region.
Introduction
Landslides rank among the most destructive natural hazards, endangering human lives, infrastructure, and the natural environment across the globe. In Iran, the northeastern part of Khuzestan Province, including the Shahid Abbaspour Dam watershed, is especially prone to landslides due to its mountainous topography, diverse geological formations, and seasonal rainfall patterns. Landslide susceptibility maps, which estimate the likelihood of occurrence based on local terrain and environmental conditions, are critical tools for planners, engineers, and policymakers. These maps guide land-use decisions, infrastructure development, and disaster preparedness, thereby reducing potential impacts on communities and ecosystems. Various methods, ranging from statistical models to machine learning techniques, have been employed to create such maps, each with varying degrees of accuracy and applicability. However, few studies have applied advanced machine learning approaches, such as Artificial Neural Networks (ANNs), in the Shahid Abbaspour Dam watershed, a critical area within the Greater Karun watershed. This research fills this gap by modeling landslide susceptibility in the region using an ANN, a method known for its ability to handle complex, non-linear relationships among multiple variables. The objective of this study is to produce a reliable susceptibility map to support risk mitigation, land-use planning, and sustainable development in this environmentally sensitive area. By leveraging the ANN’s predictive power, this study offers a robust, scientifically grounded approach to addressing a pressing regional challenge, contributing to the broader field of geohazard assessment.
Material and Methods
The study focuses on the Shahid Abbaspour Dam watershed, located in northeastern Khuzestan. This area, part of the Greater Karun watershed, features rugged terrain, diverse geology, and a history of landslide activity, making it ideal for susceptibility analysis. Our methodology was conducted in five comprehensive stages to assess the probability of landslides based on local conditions. First, we conducted an extensive literature review to identify relevant factors and gathered data from multiple sources. Second, we selected and prepared 15 influencing factors as spatial layers: elevation, slope gradient, slope aspect, curvature, Stream Power Index (SPI), Topographic Wetness Index (TWI), distance from roads, distance from rivers, proximity to faults, geological formations, soil texture, land-use patterns, precipitation, Normalized Difference Vegetation Index (NDVI), and seismic activity. These datasets were derived from 1:50,000-scale topographic maps, 1:100,000-scale geological maps, and a 30-meter resolution Digital Elevation Model (DEM) from ASTER satellite imagery. In the third stage, we implemented an Artificial Neural Network (ANN) in the R programming environment using the nnet package. This machine learning approach excels at capturing non-linear patterns among variables, making it suitable for landslide prediction. The ANN was configured with a single hidden layer of 10 neurons, trained with 70% of the dataset, and tested with the remaining 30%. The fourth stage involved generating a landslide susceptibility map, classifying the area into zones of varying risk. Finally, we validated the model using a suite of performance metrics: accuracy, Root Mean Square Error (RMSE), Kappa coefficient, Mean Absolute Error (MAE), R-squared value, sensitivity, specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV). These metrics ensured a robust evaluation of the model’s predictive accuracy and reliability, in line with best practices in geohazard modeling.
Results and Discussion
The Artificial Neural Network (ANN) model demonstrated strong predictive performance, achieving an accuracy of 0.8543, RMSE of 0.123, Kappa coefficient of 0.79, MAE of 0.098, and R-squared value of 0.9916. These metrics confirm the model’s reliability in mapping landslide susceptibility across the Shahid Abbaspour Dam watershed. Analysis revealed three primary factors driving landslide risk in the area. Proximity to rivers emerged as the most influential, likely due to mechanisms such as bank erosion, increased soil moisture, and fluctuating groundwater levels. This finding aligns with Zakerinejad and Amoshahi (2022), who noted heightened risk within 100 meters of watercourses in similar regions. Slope gradient ranked as the second key factor, with steep slopes, particularly in the southwestern Dezful region, amplifying instability. Geological formations, especially the erosion-prone Gachsaran Formation and Quaternary alluvial deposits rich in marl, ranked third, underscoring their role in susceptibility. This observation echoes Mohammadi et al. (2022), who studied the Izeh and Deh Sheikh (Abbaspour Dam) basins. In contrast, elevation and slope curvature showed limited influence, differing from Selamat et al. (2022), where elevation was a primary factor, possibly due to the unique topographic and geologic traits of our study area. Other factors, such as NDVI and seismic activity, had moderate to minor roles, likely reflecting stable vegetation cover and low seismicity. The susceptibility map classified 31.3% of the area (75.37 km²) as high-risk zones, concentrated in the southwest and a small northern section. Sensitivity analysis across susceptibility classes showed consistent performance, with values ranging from 0.8060 for the lowest class to 0.8717 for the high-susceptibility class. These results highlight the ANN’s effectiveness and its alignment with regional patterns, offering a reliable tool for understanding landslide dynamics in the watershed.
Conclusion
This study successfully developed a landslide susceptibility map for the Shahid Abbaspour Dam watershed using an Artificial Neural Network, identifying key factors and high-risk zones. Proximity to rivers, slope gradient, and geological formations, particularly the Gachsaran Formation and Quaternary alluvium, emerged as the primary drivers of landslide risk. High-susceptibility zones, covering 31.3% of the area, were predominantly located in the southwestern and northern parts, providing critical insights for targeted interventions. The ANN model performed robustly, with an accuracy of 0.8543, RMSE of 0.123, Kappa of 0.79, MAE of 0.098, and R-squared of 0.9916, validating its predictive power. These metrics underscore the model’s reliability and its potential for application in similar watersheds prone to landslides. By quantifying the influence of environmental and terrain factors, this research offers a scientific foundation for land-use planning, infrastructure design, and disaster risk management in northeastern Khuzestan. The findings can guide policymakers and planners in prioritizing mitigation efforts, such as stabilizing slopes near rivers, regulating development on steep terrain, and monitoring vulnerable geological units. Future studies could refine this model by incorporating real-time data, such as rainfall forecasts or seismic monitoring, to enhance predictive accuracy and adaptability.
Keywords
Main Subjects
)
Send comment about this article