Deep Learning-Based Assessment of Slope Creep Vulnerability Using Geophysical Survey Data

Slope creep refers to the imperceptibly slow and gradual downslope movement of soil and rock driven by gravity. It is mainly driven by moisture-induced expansion of clay-rich materials and the resulting decrease in shear strength. Although subsurface conditions can influence slope creep vulnerability, identifying their effects remains challenging. In recent years, electrical resistivity and seismic surveys have been widely used to characterize the spatial and temporal variability of subsurface soil properties. These geophysical methods provide a non-destructive means of investigating subsurface physical characteristics. In this study, electrical resistivity and seismic surveys were conducted to assess slope creep vulnerability associated with subsurface conditions. Geophysical survey data were obtained from 124 slope sites, and their slope creep vulnerability was classified into two groups (low and high) based on field investigations. Cross-plot analysis was applied to integrate electrical resistivity and seismic velocity, and the resulting data points were classified into four quadrants according to threshold values of seismic velocity and electrical resistivity. The threshold values were statistically determined using a t-test. The composition ratios of the four quadrants were used as input variables for deep learning training, and the bedrock proportion based on seismic velocity included as an additional input. As a result, a total of five input variables were used, and deep learning training was performed by classifying slope creep vulnerability into two groups. As a result, a total of five input variables were used to train a deep learning model for classification of slope creep vulnerability into two groups. Due to the limited dataset size, five-fold cross-validation was applied for model evaluation. As a result, the deep learning model achieved an accuracy of 81.5% and a recall of 83.0% in classifying slope creep vulnerability, indicating its effectiveness in identifying slope creep–prone areas. Acknowledgments: This study was carried out with the support of ´R&D Program for Forest Science Technology (RS-2025-02213451)´ provided by Korea Forest Service (Korea Forestry Promotion Institute).