Publication: Integrated artificial intelligence driven methods for evaluating earthquake-induced landslide hazards and road network vulnerability.
Loading...
Date
2025-10
Authors
Wang, Yiheng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Earthquake-induced landslides (EQIL) pose critical threats to human safety, infrastructure, and economic stability in seismically active regions such as Sichuan
Province, China. Conventional hazard assessment methods often struggle with complex environmental variables and multi-source data, limiting their scalability and predictive accuracy. This study introduces an integrated AI-GIS framework aimed at enhancing EQIL risk assessment and infrastructure vulnerability analysis. The approach combines a Python-based seismic slope stability tool, deep learning models (Temporal Convolutional Networks, Gated Recurrent Units, Random Forests), and a CNN-GIS method for evaluating road network exposure. Results indicate that slope height and soil cohesion significantly influence safety factors, with improved Sarma and Bishop methods outperforming pseudo-static models. Temporal Convolutional Networks achieved the highest predictive performance (ACC = 0.781, AUC = 0.851), while Random Forests identified slope angle, seismic intensity, and soil type as dominant susceptibility factors. The CNN-GIS model demonstrated over 90% accuracy across 8,700 road segments, enabling effective spatial risk visualization. The proposed framework offers a scalable solution for EQIL hazard analysis and supports disaster resilience planning in vulnerable regions.