Landslide Susceptibility Mapping Around Alkumru Dam (Siirt, Türkiye) Using Machine Learning and Ensemble Models

Why It Matters

Accurate susceptibility maps enable targeted disaster‑risk mitigation and protect critical infrastructure, directly supporting sustainable development in landslide‑prone regions.

Key Takeaways

• Naive Bayes achieved highest AUC of 0.959
• All models AUC ranged 0.877‑0.959
• High susceptibility zones on southern and western slopes
• 52‑76% of landslides fall in high‑risk zones
• Maps support dam safety and land‑use planning

Pulse Analysis

Machine‑learning techniques are reshaping geohazard analysis, offering faster, data‑driven alternatives to traditional statistical methods. In the Alkumru Dam study, researchers leveraged six algorithms—Support Vector Machines, Logistic Regression, Maximum Entropy, Naive Bayes, Random Forest, and Artificial Neural Networks—plus an ensemble approach to process an extensive landslide inventory and eleven conditioning factors. The high AUC values across the board demonstrate that these models can capture complex terrain‑soil‑hydrology interactions, delivering reliable susceptibility predictions that are essential for infrastructure projects situated in tectonically active zones.

Among the tested algorithms, Naive Bayes emerged as the top performer with an AUC of 0.959, likely due to its probabilistic framework handling sparse, categorical inputs efficiently. The ensemble model, which aggregates individual predictions, reinforced the spatial consistency observed across all methods, confirming that high‑risk areas are not artifacts of a single technique. Validation metrics showing that up to three‑quarters of historic landslides align with high‑ and very‑high‑risk zones underscore the practical accuracy of the approach, while the uniformity of results simplifies decision‑making for engineers and planners.

The implications extend beyond the Alkumru reservoir. By providing a scientifically vetted susceptibility map, authorities can prioritize reinforcement works, adjust reservoir operation protocols, and guide land‑use zoning to mitigate future failures. The methodology is scalable to other dam sites and mountainous catchments, especially when integrated with Geographic Information Systems and real‑time monitoring data. As climate change intensifies precipitation extremes, such predictive tools become indispensable for proactive disaster risk management and safeguarding critical water infrastructure.