Convolutional neural network-based deep learning for landslide susceptibility mapping in the Bakhtegan watershed

Area of Science:

• Geosciences
• Artificial Intelligence
• Environmental Science

Background:

• Landslides present a major risk to safety and infrastructure.
• Traditional susceptibility models struggle with complex spatial data.
• Accurate landslide assessment is vital for risk management.

Purpose of the Study:

• To develop a high-precision landslide susceptibility map using deep learning.
• To assess the effectiveness of convolutional neural networks (CNNs) in landslide prediction.
• To identify high-risk zones in the Bakhtegan watershed for targeted mitigation.

Main Methods:

• A comprehensive landslide inventory of 235 locations was created.
• Fifteen conditioning factors (topographical, geological, hydrological, climatological) were used.
• A convolutional neural network (CNN) model was trained and validated.

Main Results:

• The CNN model achieved 95.76% accuracy and 95.11% precision.
• Low error metrics (MAE, MSE, RMSE) confirm model reliability.
• Northern and northeastern Bakhtegan watershed identified as highly susceptible.

Conclusions:

• Deep learning (CNNs) offers a powerful, scalable solution for landslide susceptibility mapping.
• The study provides critical data for urban planners and policymakers.
• Proactive mitigation strategies are essential in identified high-risk areas.