AGU Fall Meeting 2024
I presented my research, “Advancing Regional Landslide Risk Assessment with Integrated AI-Physics Solutions,” in the session Landslide Life Cycle: From Hazard Analysis to Risk Assessment at the AGU Fall Meeting 2024. My work demonstrated the potential of combining AI and physics-based approaches to advance regional landslide risk assessment, drawing attention and sparking valuable discussions among attendees.
Landslides are increasing in frequency, duration, and severity due to climate change, global population growth, and urbanization. Even more concerning is the rise of multiple-occurrence landslides triggered by single rainfall events. This underscores the urgent need for regional-scale landslide susceptibility and risk assessments to identify areas with high susceptibility and risk.
To address this need, my study makes several contributions:
-
Few studies have combined the strengths of data-driven and physics-based models for susceptibility mapping. I proposed an AI–physics integrated spatial susceptibility mapping model that incorporates infinite slope stability analysis and logistic regression.
-
I introduced a pixel-to-slope transformation method, where slope units are extracted using hydrological analysis and AI algorithms. Statistical parameters—such as the mean and maximum—of pixel-level susceptibility within each slope unit are computed to characterize slope-scale susceptibility.
-
To mitigate potential accuracy loss during transformation, I proposed a calibration method based on the standard deviation of susceptibility. Field verification is recommended when the within-unit standard deviation exceeds a certain threshold.
-
I conducted quantitative analyses of population and economic vulnerabilities and proposed a comprehensive vulnerability assessment matrix. This, combined with susceptibility, forms a regional risk assessment matrix.
-
I developed a deep learning method for temporal susceptibility prediction that integrates convolutional neural networks and recurrent neural networks. Typical landslide-prone regions in the Three Gorges Reservoir area of China were used as the study site.
The findings show that the proposed integrated AI-physics approach significantly improves the accuracy, generalizability, and practicality of landslide susceptibility and risk assessment. This study provides an easy-to-implement and effective framework for regional landslide risk assessment.
