Year: 2025
Type: article
Abstract: In areas with vulnerable terrain, landslides pose a significant risk to both lives and infrastructure, making precise susceptibility maps essential for effective risk management and planning. This research introduces an advanced Deep Learning (DL) framework that combines multiple models, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Attention U-Net, and their combinations (CNN + LSTM and CNN + Attention), to enhance the landslide susceptibility prediction. The models were trained and validated using 1,580 landslide occurrence data points from the Western Ghats of Karnataka, with 17 geospatial conditioning factors derived from satellite remote sensing and processed through Google Earth Engine (GEE). To verify the reliability of the input features, multicollinearity was evaluated using VIF and TOL. A comparative analysis showed that the LSTM model achieved the highest accuracy at 98.80% and an Area Under the Curve (AUC) of 0.988, with CNN + Attention U-Net performing similarly. The CNN + LSTM hybrid model demonstrated an exceptional spatial detection ability, identifying the highest proportion of landslide risk areas. The generated susceptibility map offers a reliable, data-driven resource for early warning systems, land-use planning, and disaster resilience strategies. This map provides practical insights for the government agencies, urban planners, and disaster response teams to develop proactive hazard mitigation measures.
DOI: 10.48084/etasr.11505
Authors: D. Anil, S. H. Manjula
Institution: Bangalore University
Language: English
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
SDG: Life on land
Open Access status: gold
Estimated APC paid: $431
OpenAlex ID: W4412880378