Utilizing artificial intelligence techniques for soil depth prediction and its influences in landslide hazard modeling

Abstract

Soil depth plays a pivotal role in determining hillslope stability, understanding hydrogeology, promoting optimal vegetation growth, and comprehensively elucidating soil erosion dynamics. In this study, two robust artificial intelligence methodologies, quantile regression forest (QRF) and deep neural network (DNN), were employed to predict spatial variations in soil depth across a digital terrain. Particularly during periods of intense rainfall, shallow landslides pose recurrent threats to human safety and property integrity. Thus, the identification of potential landslide-prone regions becomes imperative for mitigating associated risks. During slope stability analyses, soil depth assumes significance; nonetheless, data regarding soil depth from areas prone to landslides are rarely obtained. The main objective of this study is to explore the impact of incorporating soil depth spatial distributions on the predictive capabilities of shallow landslide model within a given terrain. By leveraging two distinct spatial soil depth distributions, a comprehensive analysis of slope stability analysis was conducted. The significance of soil depth spatial distribution, particularly when employing DNN-generated data, is underscored in refining predictions and preventing overestimations of landslide-prone or stable regions. Notably, integration of DNN-derived soil depth data into the infinite slope model yielded a marked enhancement in the accuracy of factor of safety (FS) distributions, achieving an impressive 86.9% accuracy rate while QRF-derived FS has shown 74.7% accuracy. This analytical approach, while straightforward, offers a powerful tool for evaluating slope instability and forecasting shallow landslides, thereby facilitating proactive mitigation measures.